Migrations
/
pytorch-image-models
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '67d0665b46'
${ noResults }
pytorch-image-models/timm/models/efficientnet.py

1789 lines
72 KiB
Raw Normal View History Unescape

Update EfficientNet comments
5 years ago
""" PyTorch EfficientNet Family
An implementation of EfficienNet that covers variety of related models with efficient architectures:
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
* EfficientNet (B0-B8, L2 + Tensorflow pretrained AutoAug/RandAug/AdvProp/NoisyStudent weight ports)
Update EfficientNet comments
5 years ago
- EfficientNet: Rethinking Model Scaling for CNNs - https://arxiv.org/abs/1905.11946
- CondConv: Conditionally Parameterized Convolutions for Efficient Inference - https://arxiv.org/abs/1904.04971
- Adversarial Examples Improve Image Recognition - https://arxiv.org/abs/1911.09665
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
- Self-training with Noisy Student improves ImageNet classification - https://arxiv.org/abs/1911.04252
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
* MixNet (Small, Medium, and Large)
Update EfficientNet comments
5 years ago
- MixConv: Mixed Depthwise Convolutional Kernels - https://arxiv.org/abs/1907.09595
* MNasNet B1, A1 (SE), Small
- MnasNet: Platform-Aware Neural Architecture Search for Mobile - https://arxiv.org/abs/1807.11626
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
* FBNet-C
Update EfficientNet comments
5 years ago
- FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable NAS - https://arxiv.org/abs/1812.03443
Add Single-Path NAS pixel1 model
6 years ago
* Single-Path NAS Pixel1
Update EfficientNet comments
5 years ago
- Single-Path NAS: Designing Hardware-Efficient ConvNets - https://arxiv.org/abs/1904.02877
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
Update EfficientNet comments
5 years ago
* And likely more...
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
Fix some attributions, add copyrights to some file docstrings
4 years ago
Hacked together by / Copyright 2020 Ross Wightman
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
"""
Add missing feature_info() on MobileNetV3, make hook feature output order/type consistent with bottleneck (list, decreasing fmap size)
5 years ago
import torch
Monster commit, activation refactor, VoVNet, norm_act improvements, more * refactor activations into basic PyTorch, jit scripted, and memory efficient custom auto * implement hard-mish, better grad for hard-swish * add initial VovNet V1/V2 impl, fix #151 * VovNet and DenseNet first models to use NormAct layers (support BatchNormAct2d, EvoNorm, InplaceIABN) * Wrap IABN for any models that use it * make more models torchscript compatible (DPN, PNasNet, Res2Net, SelecSLS) and add tests
5 years ago
import torch.nn as nn
import torch.nn.functional as F
Add missing feature_info() on MobileNetV3, make hook feature output order/type consistent with bottleneck (list, decreasing fmap size)
5 years ago
from typing import List
Replace all None by nn.Identity() in all models reset_classifier when False-values num_classes is given. Make small code refactoring
5 years ago
from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
Monster commit, activation refactor, VoVNet, norm_act improvements, more * refactor activations into basic PyTorch, jit scripted, and memory efficient custom auto * implement hard-mish, better grad for hard-swish * add initial VovNet V1/V2 impl, fix #151 * VovNet and DenseNet first models to use NormAct layers (support BatchNormAct2d, EvoNorm, InplaceIABN) * Wrap IABN for any models that use it * make more models torchscript compatible (DPN, PNasNet, Res2Net, SelecSLS) and add tests
5 years ago
from .efficientnet_blocks import round_channels, resolve_bn_args, resolve_act_layer, BN_EPS_TF_DEFAULT
from .efficientnet_builder import EfficientNetBuilder, decode_arch_def, efficientnet_init_weights
Move FeatureHooks into features.py, switch EfficientNet, MobileNetV3 to use build model helper
4 years ago
from .features import FeatureInfo, FeatureHooks
Add default_cfg back to models wrapped in feature extraction module as per discussion in #294.
4 years ago
from .helpers import build_model_with_cfg, default_cfg_for_features
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
from .layers import create_conv2d, create_classifier
Replace all None by nn.Identity() in all models reset_classifier when False-values num_classes is given. Make small code refactoring
5 years ago
from .registry import register_model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
__all__ = ['EfficientNet']
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
def _cfg(url='', **kwargs):
return {
'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7),
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
'crop_pct': 0.875, 'interpolation': 'bicubic',
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
'first_conv': 'conv_stem', 'classifier': 'classifier',
**kwargs
}
default_cfgs = {
Add MobileNetV3 and associated changes hard-swish, hard-sigmoid, efficient head, etc
6 years ago
'mnasnet_050': _cfg(url=''),
'mnasnet_075': _cfg(url=''),
Add the MnasNet-B1 variant weights, add A1/B1 model names as in the stand-alone repo, remove a bit of unused code
6 years ago
'mnasnet_100': _cfg(
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mnasnet_b1-74cb7081.pth'),
Add MobileNetV3 and associated changes hard-swish, hard-sigmoid, efficient head, etc
6 years ago
'mnasnet_140': _cfg(url=''),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add MobileNetV3 and associated changes hard-swish, hard-sigmoid, efficient head, etc
6 years ago
'semnasnet_050': _cfg(url=''),
'semnasnet_075': _cfg(url=''),
Add native PyTorch weights for SE-MnasNet aka MnasNet-A1
6 years ago
'semnasnet_100': _cfg(
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mnasnet_a1-d9418771.pth'),
Add MobileNetV3 and associated changes hard-swish, hard-sigmoid, efficient head, etc
6 years ago
'semnasnet_140': _cfg(url=''),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
'mnasnet_small': _cfg(url=''),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add final weights for MobileNet-V2 experiments
5 years ago
'mobilenetv2_100': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mobilenetv2_100_ra-b33bc2c4.pth'),
'mobilenetv2_110d': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mobilenetv2_110d_ra-77090ade.pth'),
'mobilenetv2_120d': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mobilenetv2_120d_ra-5987e2ed.pth'),
'mobilenetv2_140': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mobilenetv2_140_ra-21a4e913.pth'),
Experimenting with some MobileNetV2 variations to compare against EfficientNet-Lite
5 years ago
PyTorch trained MobileNetV3 weights that match/best paper. Update fbnetc weights to improved copy.
6 years ago
'fbnetc_100': _cfg(
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/fbnetc_100-c345b898.pth',
interpolation='bilinear'),
PyTorch trained MobileNetV3 weights that match/best paper. Update fbnetc weights to improved copy.
6 years ago
'spnasnet_100': _cfg(
Update version, results csv files, and move remaining dropbox weights to github
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/spnasnet_100-048bc3f4.pth',
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
interpolation='bilinear'),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add native PyTorch weights for EfficientNet-B0 w/ top-1 > 76.9 * also add pooling details to default cfg for efficiennets so testtimepool wrapper works
6 years ago
'efficientnet_b0': _cfg(
Add updated RandAugment trained EfficientNet-B0 trained weights from @michaelklachko
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b0_ra-3dd342df.pth'),
Add native PyTorch weights for EfficientNet-B0 w/ top-1 > 76.9 * also add pooling details to default cfg for efficiennets so testtimepool wrapper works
6 years ago
'efficientnet_b1': _cfg(
Add native PyTorch EfficientNet B1 and B2 weights. Not quite where I want them, but hitting the brick wall and moving on to other projects...
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b1-533bc792.pth',
Add updated EfficientNet-V2S weights, 83.8 @ 384x384 test. Add PyTorch trained EfficientNet-B4 weights, 83.4 @ 384x384 test. Tweak non TF EfficientNet B1-B4 train/test res scaling.
4 years ago
test_input_size=(3, 256, 256), crop_pct=1.0),
Add native PyTorch weights for EfficientNet-B0 w/ top-1 > 76.9 * also add pooling details to default cfg for efficiennets so testtimepool wrapper works
6 years ago
'efficientnet_b2': _cfg(
New PyTorch trained EfficientNet-B2 weights with my RandAugment impl
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b2_ra-bcdf34b7.pth',
Add updated EfficientNet-V2S weights, 83.8 @ 384x384 test. Add PyTorch trained EfficientNet-B4 weights, 83.4 @ 384x384 test. Tweak non TF EfficientNet B1-B4 train/test res scaling.
4 years ago
input_size=(3, 256, 256), pool_size=(8, 8), test_input_size=(3, 288, 288), crop_pct=1.0),
Add native PyTorch weights for EfficientNet-B0 w/ top-1 > 76.9 * also add pooling details to default cfg for efficiennets so testtimepool wrapper works
6 years ago
'efficientnet_b3': _cfg(
Add new EfficientNet-B3 and RegNetY-3.2GF weights, both just over 82 top-1
4 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b3_ra2-cf984f9c.pth',
Add updated EfficientNet-V2S weights, 83.8 @ 384x384 test. Add PyTorch trained EfficientNet-B4 weights, 83.4 @ 384x384 test. Tweak non TF EfficientNet B1-B4 train/test res scaling.
4 years ago
input_size=(3, 288, 288), pool_size=(9, 9), test_input_size=(3, 320, 320), crop_pct=1.0),
Add native PyTorch weights for EfficientNet-B0 w/ top-1 > 76.9 * also add pooling details to default cfg for efficiennets so testtimepool wrapper works
6 years ago
'efficientnet_b4': _cfg(
Add updated EfficientNet-V2S weights, 83.8 @ 384x384 test. Add PyTorch trained EfficientNet-B4 weights, 83.4 @ 384x384 test. Tweak non TF EfficientNet B1-B4 train/test res scaling.
4 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b4_ra2_320-7eb33cd5.pth',
input_size=(3, 320, 320), pool_size=(10, 10), test_input_size=(3, 384, 384), crop_pct=1.0),
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
'efficientnet_b5': _cfg(
Add native PyTorch EfficientNet B1 and B2 weights. Not quite where I want them, but hitting the brick wall and moving on to other projects...
5 years ago
url='', input_size=(3, 456, 456), pool_size=(15, 15), crop_pct=0.934),
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
'efficientnet_b6': _cfg(
url='', input_size=(3, 528, 528), pool_size=(17, 17), crop_pct=0.942),
'efficientnet_b7': _cfg(
url='', input_size=(3, 600, 600), pool_size=(19, 19), crop_pct=0.949),
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
'efficientnet_b8': _cfg(
url='', input_size=(3, 672, 672), pool_size=(21, 21), crop_pct=0.954),
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
'efficientnet_l2': _cfg(
url='', input_size=(3, 800, 800), pool_size=(25, 25), crop_pct=0.961),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
'efficientnet_es': _cfg(
Add PyTorch trained EfficientNet-ES weights from Andrew Lavin
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_es_ra-f111e99c.pth'),
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
'efficientnet_em': _cfg(
Add EfficientNet-EdgeTPU-M (efficientnet_em) model trained natively in PyTorch. More sotabench fiddling.
4 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_em_ra2-66250f76.pth',
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
'efficientnet_el': _cfg(
adding efficientnet_el, efficientnet_es_prunedefficientnet_es_pruned and efficientnet_el_pruned pretrained models to timm/models
4 years ago
url='https://github.com/DeGirum/pruned-models/releases/download/efficientnet_v1.0/efficientnet_el.pth',
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904),
'efficientnet_es_pruned': _cfg(
url='https://github.com/DeGirum/pruned-models/releases/download/efficientnet_v1.0/efficientnet_es_pruned75.pth'),
'efficientnet_el_pruned': _cfg(
url='https://github.com/DeGirum/pruned-models/releases/download/efficientnet_v1.0/efficientnet_el_pruned70.pth',
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_cc_b0_4e': _cfg(url=''),
'efficientnet_cc_b0_8e': _cfg(url=''),
'efficientnet_cc_b1_8e': _cfg(url='', input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
'efficientnet_lite0': _cfg(
Add EfficientNet-Lite0 weights trained with this code by @hal-314, 75.484 top-1
4 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_lite0_ra-37913777.pth'),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
'efficientnet_lite1': _cfg(
url='',
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
'efficientnet_lite2': _cfg(
url='',
input_size=(3, 260, 260), pool_size=(9, 9), crop_pct=0.890),
'efficientnet_lite3': _cfg(
url='',
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904),
'efficientnet_lite4': _cfg(
url='', input_size=(3, 380, 380), pool_size=(12, 12), crop_pct=0.922),
added efficientnet pruned weights
5 years ago
'efficientnet_b1_pruned': _cfg(
url='https://imvl-automl-sh.oss-cn-shanghai.aliyuncs.com/darts/hyperml/hyperml/job_45403/outputs/effnetb1_pruned_9ebb3fe6.pth',
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882, mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD),
'efficientnet_b2_pruned': _cfg(
url='https://imvl-automl-sh.oss-cn-shanghai.aliyuncs.com/darts/hyperml/hyperml/job_45403/outputs/effnetb2_pruned_203f55bc.pth',
input_size=(3, 260, 260), pool_size=(9, 9), crop_pct=0.890, mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD),
'efficientnet_b3_pruned': _cfg(
url='https://imvl-automl-sh.oss-cn-shanghai.aliyuncs.com/darts/hyperml/hyperml/job_45403/outputs/effnetb3_pruned_5abcc29f.pth',
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904, mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD),
EfficientNet-V2S preliminary model def (for experimentation)
4 years ago
'efficientnet_v2s': _cfg(
Add updated EfficientNet-V2S weights, 83.8 @ 384x384 test. Add PyTorch trained EfficientNet-B4 weights, 83.4 @ 384x384 test. Tweak non TF EfficientNet B1-B4 train/test res scaling.
4 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_v2s_ra2_288-a6477665.pth',
input_size=(3, 288, 288), test_input_size=(3, 384, 384), pool_size=(9, 9), crop_pct=1.0), # FIXME WIP
EfficientNet-V2S preliminary model def (for experimentation)
4 years ago
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
'tf_efficientnet_b0': _cfg(
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b0_aa-827b6e33.pth',
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
input_size=(3, 224, 224)),
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
'tf_efficientnet_b1': _cfg(
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b1_aa-ea7a6ee0.pth',
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
'tf_efficientnet_b2': _cfg(
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b2_aa-60c94f97.pth',
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
input_size=(3, 260, 260), pool_size=(9, 9), crop_pct=0.890),
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
'tf_efficientnet_b3': _cfg(
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b3_aa-84b4657e.pth',
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904),
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
'tf_efficientnet_b4': _cfg(
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b4_aa-818f208c.pth',
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
input_size=(3, 380, 380), pool_size=(12, 12), crop_pct=0.922),
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
'tf_efficientnet_b5': _cfg(
Add TF RandAug weights for B5/B7 EfficientNet models.
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b5_ra-9a3e5369.pth',
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
input_size=(3, 456, 456), pool_size=(15, 15), crop_pct=0.934),
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
'tf_efficientnet_b6': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b6_aa-80ba17e4.pth',
input_size=(3, 528, 528), pool_size=(17, 17), crop_pct=0.942),
'tf_efficientnet_b7': _cfg(
Add TF RandAug weights for B5/B7 EfficientNet models.
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b7_ra-6c08e654.pth',
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
input_size=(3, 600, 600), pool_size=(19, 19), crop_pct=0.949),
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
'tf_efficientnet_b8': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b8_ra-572d5dd9.pth',
input_size=(3, 672, 672), pool_size=(21, 21), crop_pct=0.954),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
'tf_efficientnet_b0_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b0_ap-f262efe1.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD, input_size=(3, 224, 224)),
'tf_efficientnet_b1_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b1_ap-44ef0a3d.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
'tf_efficientnet_b2_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b2_ap-2f8e7636.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 260, 260), pool_size=(9, 9), crop_pct=0.890),
'tf_efficientnet_b3_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b3_ap-aad25bdd.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904),
'tf_efficientnet_b4_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b4_ap-dedb23e6.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 380, 380), pool_size=(12, 12), crop_pct=0.922),
'tf_efficientnet_b5_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b5_ap-9e82fae8.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 456, 456), pool_size=(15, 15), crop_pct=0.934),
'tf_efficientnet_b6_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b6_ap-4ffb161f.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 528, 528), pool_size=(17, 17), crop_pct=0.942),
'tf_efficientnet_b7_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b7_ap-ddb28fec.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 600, 600), pool_size=(19, 19), crop_pct=0.949),
'tf_efficientnet_b8_ap': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b8_ap-00e169fa.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 672, 672), pool_size=(21, 21), crop_pct=0.954),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
'tf_efficientnet_b0_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b0_ns-c0e6a31c.pth',
input_size=(3, 224, 224)),
'tf_efficientnet_b1_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b1_ns-99dd0c41.pth',
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
'tf_efficientnet_b2_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b2_ns-00306e48.pth',
input_size=(3, 260, 260), pool_size=(9, 9), crop_pct=0.890),
'tf_efficientnet_b3_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b3_ns-9d44bf68.pth',
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904),
'tf_efficientnet_b4_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b4_ns-d6313a46.pth',
input_size=(3, 380, 380), pool_size=(12, 12), crop_pct=0.922),
'tf_efficientnet_b5_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b5_ns-6f26d0cf.pth',
input_size=(3, 456, 456), pool_size=(15, 15), crop_pct=0.934),
'tf_efficientnet_b6_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b6_ns-51548356.pth',
input_size=(3, 528, 528), pool_size=(17, 17), crop_pct=0.942),
'tf_efficientnet_b7_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b7_ns-1dbc32de.pth',
input_size=(3, 600, 600), pool_size=(19, 19), crop_pct=0.949),
'tf_efficientnet_l2_ns_475': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_l2_ns_475-bebbd00a.pth',
input_size=(3, 475, 475), pool_size=(15, 15), crop_pct=0.936),
'tf_efficientnet_l2_ns': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_l2_ns-df73bb44.pth',
Batch validation batch size adjustment, tweak L2 crop pct
5 years ago
input_size=(3, 800, 800), pool_size=(25, 25), crop_pct=0.96),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
'tf_efficientnet_es': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_es-ca1afbfe.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 224, 224), ),
'tf_efficientnet_em': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_em-e78cfe58.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
'tf_efficientnet_el': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_el-5143854e.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_cc_b0_4e': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_cc_b0_4e-4362b6b2.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD),
'tf_efficientnet_cc_b0_8e': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_cc_b0_8e-66184a25.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD),
'tf_efficientnet_cc_b1_8e': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_cc_b1_8e-f7c79ae1.pth',
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
'tf_efficientnet_lite0': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_lite0-0aa007d2.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
interpolation='bicubic', # should be bilinear but bicubic better match for TF bilinear at low res
),
'tf_efficientnet_lite1': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_lite1-bde8b488.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 240, 240), pool_size=(8, 8), crop_pct=0.882,
interpolation='bicubic', # should be bilinear but bicubic better match for TF bilinear at low res
),
'tf_efficientnet_lite2': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_lite2-dcccb7df.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 260, 260), pool_size=(9, 9), crop_pct=0.890,
interpolation='bicubic', # should be bilinear but bicubic better match for TF bilinear at low res
),
'tf_efficientnet_lite3': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_lite3-b733e338.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 300, 300), pool_size=(10, 10), crop_pct=0.904, interpolation='bilinear'),
'tf_efficientnet_lite4': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_lite4-741542c3.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
Add EfficientNet-Lite results, update README
5 years ago
input_size=(3, 380, 380), pool_size=(12, 12), crop_pct=0.920, interpolation='bilinear'),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add MixNet Small and Large PyTorch native weights (no same padding)
5 years ago
'mixnet_s': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_s-a907afbc.pth'),
Add native PyTorch weights for MixNet-Medium with no SAME padding necessary. Remove unused block of code.
5 years ago
'mixnet_m': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_m-4647fc68.pth'),
Add MixNet Small and Large PyTorch native weights (no same padding)
5 years ago
'mixnet_l': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_l-5a9a2ed8.pth'),
Add weights for my MixNet-XL creation, include README updates for EdgeTPU models
5 years ago
'mixnet_xl': _cfg(
Add update RandAugment MixNet-XL weights
5 years ago
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_xl_ra-aac3c00c.pth'),
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
'mixnet_xxl': _cfg(),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
'tf_mixnet_s': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mixnet_s-89d3354b.pth'),
'tf_mixnet_m': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mixnet_m-0f4d8805.pth'),
'tf_mixnet_l': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mixnet_l-6c92e0c8.pth'),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
}
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
_DEBUG = False
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
Add MobileNetV3 and associated changes hard-swish, hard-sigmoid, efficient head, etc
6 years ago
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
class EfficientNet(nn.Module):
""" (Generic) EfficientNet
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
A flexible and performant PyTorch implementation of efficient network architectures, including:
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
* EfficientNet B0-B8, L2
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
* EfficientNet-EdgeTPU
* EfficientNet-CondConv
* MixNet S, M, L, XL
* MnasNet A1, B1, and small
* FBNet C
* Single-Path NAS Pixel1
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
"""
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
def __init__(self, block_args, num_classes=1000, num_features=1280, in_chans=3, stem_size=32,
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
channel_multiplier=1.0, channel_divisor=8, channel_min=None,
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
output_stride=32, pad_type='', fix_stem=False, act_layer=nn.ReLU, drop_rate=0., drop_path_rate=0.,
Finish with HRNet, weights and models updated. Improve consistency in model classifier/global pool treatment.
5 years ago
se_kwargs=None, norm_layer=nn.BatchNorm2d, norm_kwargs=None, global_pool='avg'):
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
super(EfficientNet, self).__init__()
norm_kwargs = norm_kwargs or {}
self.num_classes = num_classes
self.num_features = num_features
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
self.drop_rate = drop_rate
# Stem
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
if not fix_stem:
stem_size = round_channels(stem_size, channel_multiplier, channel_divisor, channel_min)
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
self.conv_stem = create_conv2d(in_chans, stem_size, 3, stride=2, padding=pad_type)
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
self.bn1 = norm_layer(stem_size, **norm_kwargs)
self.act1 = act_layer(inplace=True)
# Middle stages (IR/ER/DS Blocks)
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
builder = EfficientNetBuilder(
Layer refactoring continues, ResNet downsample rewrite for proper dilation in 3x3 and avg_pool cases * select_conv2d -> create_conv2d * added create_attn to create attention module from string/bool/module * factor padding helpers into own file, use in both conv2d_same and avg_pool2d_same * add some more test eca resnet variants * minor tweaks, naming, comments, consistency
5 years ago
channel_multiplier, channel_divisor, channel_min, output_stride, pad_type, act_layer, se_kwargs,
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
norm_layer, norm_kwargs, drop_path_rate, verbose=_DEBUG)
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
self.blocks = nn.Sequential(*builder(stem_size, block_args))
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
self.feature_info = builder.features
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
head_chs = builder.in_chs
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
# Head + Pooling
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
self.conv_head = create_conv2d(head_chs, self.num_features, 1, padding=pad_type)
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
self.bn2 = norm_layer(self.num_features, **norm_kwargs)
self.act2 = act_layer(inplace=True)
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
self.global_pool, self.classifier = create_classifier(
self.num_features, self.num_classes, pool_type=global_pool)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
Finish with HRNet, weights and models updated. Improve consistency in model classifier/global pool treatment.
5 years ago
efficientnet_init_weights(self)
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def as_sequential(self):
layers = [self.conv_stem, self.bn1, self.act1]
layers.extend(self.blocks)
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
layers.extend([self.conv_head, self.bn2, self.act2, self.global_pool])
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
layers.extend([nn.Dropout(self.drop_rate), self.classifier])
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
return nn.Sequential(*layers)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
def get_classifier(self):
return self.classifier
def reset_classifier(self, num_classes, global_pool='avg'):
self.num_classes = num_classes
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
self.global_pool, self.classifier = create_classifier(
self.num_features, self.num_classes, pool_type=global_pool)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def forward_features(self, x):
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
x = self.conv_stem(x)
x = self.bn1(x)
x = self.act1(x)
x = self.blocks(x)
x = self.conv_head(x)
x = self.bn2(x)
x = self.act2(x)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return x
def forward(self, x):
x = self.forward_features(x)
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
x = self.global_pool(x)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
if self.drop_rate > 0.:
x = F.dropout(x, p=self.drop_rate, training=self.training)
return self.classifier(x)
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
class EfficientNetFeatures(nn.Module):
""" EfficientNet Feature Extractor
A work-in-progress feature extraction module for EfficientNet, to use as a backbone for segmentation
and object detection models.
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
"""
Update EfficientNet feature extraction for EfficientDet. Add needed MaxPoolSame as well.
5 years ago
def __init__(self, block_args, out_indices=(0, 1, 2, 3, 4), feature_location='bottleneck',
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
in_chans=3, stem_size=32, channel_multiplier=1.0, channel_divisor=8, channel_min=None,
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
output_stride=32, pad_type='', fix_stem=False, act_layer=nn.ReLU, drop_rate=0., drop_path_rate=0.,
Finish with HRNet, weights and models updated. Improve consistency in model classifier/global pool treatment.
5 years ago
se_kwargs=None, norm_layer=nn.BatchNorm2d, norm_kwargs=None):
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
super(EfficientNetFeatures, self).__init__()
norm_kwargs = norm_kwargs or {}
self.drop_rate = drop_rate
# Stem
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
if not fix_stem:
stem_size = round_channels(stem_size, channel_multiplier, channel_divisor, channel_min)
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
self.conv_stem = create_conv2d(in_chans, stem_size, 3, stride=2, padding=pad_type)
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
self.bn1 = norm_layer(stem_size, **norm_kwargs)
self.act1 = act_layer(inplace=True)
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
# Middle stages (IR/ER/DS Blocks)
builder = EfficientNetBuilder(
channel_multiplier, channel_divisor, channel_min, output_stride, pad_type, act_layer, se_kwargs,
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
norm_layer, norm_kwargs, drop_path_rate, feature_location=feature_location, verbose=_DEBUG)
More uniform treatment of classifiers across all models, reduce code duplication.
4 years ago
self.blocks = nn.Sequential(*builder(stem_size, block_args))
Working on feature extraction, interfaces refined, a number of models working, some in progress.
4 years ago
self.feature_info = FeatureInfo(builder.features, out_indices)
Cleanup EfficientNet/MobileNetV3 feature extraction a bit, only two tap locations now, small mobilenetv3 models work
4 years ago
self._stage_out_idx = {v['stage']: i for i, v in enumerate(self.feature_info) if i in out_indices}
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
Finish with HRNet, weights and models updated. Improve consistency in model classifier/global pool treatment.
5 years ago
efficientnet_init_weights(self)
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
# Register feature extraction hooks with FeatureHooks helper
Update EfficientNet feature extraction for EfficientDet. Add needed MaxPoolSame as well.
5 years ago
self.feature_hooks = None
if feature_location != 'bottleneck':
Cleanup FeatureInfo getters, add TF models sourced Xception41/65/71 weights
4 years ago
hooks = self.feature_info.get_dicts(keys=('module', 'hook_type'))
Update EfficientNet feature extraction for EfficientDet. Add needed MaxPoolSame as well.
5 years ago
self.feature_hooks = FeatureHooks(hooks, self.named_modules())
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
Add missing feature_info() on MobileNetV3, make hook feature output order/type consistent with bottleneck (list, decreasing fmap size)
5 years ago
def forward(self, x) -> List[torch.Tensor]:
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
x = self.conv_stem(x)
x = self.bn1(x)
x = self.act1(x)
Update EfficientNet feature extraction for EfficientDet. Add needed MaxPoolSame as well.
5 years ago
if self.feature_hooks is None:
features = []
Cleanup EfficientNet/MobileNetV3 feature extraction a bit, only two tap locations now, small mobilenetv3 models work
4 years ago
if 0 in self._stage_out_idx:
features.append(x) # add stem out
Update EfficientNet feature extraction for EfficientDet. Add needed MaxPoolSame as well.
5 years ago
for i, b in enumerate(self.blocks):
x = b(x)
Cleanup EfficientNet/MobileNetV3 feature extraction a bit, only two tap locations now, small mobilenetv3 models work
4 years ago
if i + 1 in self._stage_out_idx:
Update EfficientNet feature extraction for EfficientDet. Add needed MaxPoolSame as well.
5 years ago
features.append(x)
return features
else:
self.blocks(x)
Cleanup EfficientNet/MobileNetV3 feature extraction a bit, only two tap locations now, small mobilenetv3 models work
4 years ago
out = self.feature_hooks.get_output(x.device)
return list(out.values())
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
def _create_effnet(variant, pretrained=False, **kwargs):
Bug in last mod for features_only default_cfg
4 years ago
features_only = False
model_cls = EfficientNet
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
kwargs_filter = None
if kwargs.pop('features_only', False):
Bug in last mod for features_only default_cfg
4 years ago
features_only = True
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
kwargs_filter = ('num_classes', 'num_features', 'head_conv', 'global_pool')
Move FeatureHooks into features.py, switch EfficientNet, MobileNetV3 to use build model helper
4 years ago
model_cls = EfficientNetFeatures
Add default_cfg back to models wrapped in feature extraction module as per discussion in #294.
4 years ago
model = build_model_with_cfg(
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model_cls, variant, pretrained,
default_cfg=default_cfgs[variant],
pretrained_strict=not features_only,
kwargs_filter=kwargs_filter,
**kwargs)
Bug in last mod for features_only default_cfg
4 years ago
if features_only:
model.default_cfg = default_cfg_for_features(model.default_cfg)
Add default_cfg back to models wrapped in feature extraction module as per discussion in #294.
4 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_mnasnet_a1(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
"""Creates a mnasnet-a1 model.
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet
Paper: https://arxiv.org/pdf/1807.11626.pdf.
Args:
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier: multiplier to number of channels per layer.
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
"""
arch_def = [
# stage 0, 112x112 in
['ds_r1_k3_s1_e1_c16_noskip'],
# stage 1, 112x112 in
['ir_r2_k3_s2_e6_c24'],
# stage 2, 56x56 in
['ir_r3_k5_s2_e3_c40_se0.25'],
# stage 3, 28x28 in
['ir_r4_k3_s2_e6_c80'],
# stage 4, 14x14in
['ir_r2_k3_s1_e6_c112_se0.25'],
# stage 5, 14x14in
['ir_r3_k5_s2_e6_c160_se0.25'],
# stage 6, 7x7 in
['ir_r1_k3_s1_e6_c320'],
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
stem_size=32,
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
**kwargs
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_mnasnet_b1(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
"""Creates a mnasnet-b1 model.
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet
Paper: https://arxiv.org/pdf/1807.11626.pdf.
Args:
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier: multiplier to number of channels per layer.
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
"""
arch_def = [
# stage 0, 112x112 in
['ds_r1_k3_s1_c16_noskip'],
# stage 1, 112x112 in
['ir_r3_k3_s2_e3_c24'],
# stage 2, 56x56 in
['ir_r3_k5_s2_e3_c40'],
# stage 3, 28x28 in
['ir_r3_k5_s2_e6_c80'],
# stage 4, 14x14in
['ir_r2_k3_s1_e6_c96'],
# stage 5, 14x14in
['ir_r4_k5_s2_e6_c192'],
# stage 6, 7x7 in
['ir_r1_k3_s1_e6_c320_noskip']
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
stem_size=32,
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
**kwargs
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_mnasnet_small(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
"""Creates a mnasnet-b1 model.
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet
Paper: https://arxiv.org/pdf/1807.11626.pdf.
Args:
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier: multiplier to number of channels per layer.
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
"""
arch_def = [
['ds_r1_k3_s1_c8'],
['ir_r1_k3_s2_e3_c16'],
['ir_r2_k3_s2_e6_c16'],
['ir_r4_k5_s2_e6_c32_se0.25'],
['ir_r3_k3_s1_e6_c32_se0.25'],
['ir_r3_k5_s2_e6_c88_se0.25'],
['ir_r1_k3_s1_e6_c144']
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
stem_size=8,
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
**kwargs
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Fiddling...
5 years ago
def _gen_mobilenet_v2(
variant, channel_multiplier=1.0, depth_multiplier=1.0, fix_stem_head=False, pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" Generate MobileNet-V2 network
Ref impl: https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet_v2.py
Paper: https://arxiv.org/abs/1801.04381
"""
arch_def = [
['ds_r1_k3_s1_c16'],
['ir_r2_k3_s2_e6_c24'],
['ir_r3_k3_s2_e6_c32'],
['ir_r4_k3_s2_e6_c64'],
['ir_r3_k3_s1_e6_c96'],
['ir_r3_k3_s2_e6_c160'],
['ir_r1_k3_s1_e6_c320'],
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Fiddling...
5 years ago
block_args=decode_arch_def(arch_def, depth_multiplier=depth_multiplier, fix_first_last=fix_stem_head),
num_features=1280 if fix_stem_head else round_channels(1280, channel_multiplier, 8, None),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
stem_size=32,
Fiddling...
5 years ago
fix_stem=fix_stem_head,
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Monster commit, activation refactor, VoVNet, norm_act improvements, more * refactor activations into basic PyTorch, jit scripted, and memory efficient custom auto * implement hard-mish, better grad for hard-swish * add initial VovNet V1/V2 impl, fix #151 * VovNet and DenseNet first models to use NormAct layers (support BatchNormAct2d, EvoNorm, InplaceIABN) * Wrap IABN for any models that use it * make more models torchscript compatible (DPN, PNasNet, Res2Net, SelecSLS) and add tests
5 years ago
act_layer=resolve_act_layer(kwargs, 'relu6'),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
**kwargs
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_fbnetc(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" FBNet-C
Paper: https://arxiv.org/abs/1812.03443
Ref Impl: https://github.com/facebookresearch/maskrcnn-benchmark/blob/master/maskrcnn_benchmark/modeling/backbone/fbnet_modeldef.py
NOTE: the impl above does not relate to the 'C' variant here, that was derived from paper,
it was used to confirm some building block details
"""
arch_def = [
['ir_r1_k3_s1_e1_c16'],
['ir_r1_k3_s2_e6_c24', 'ir_r2_k3_s1_e1_c24'],
['ir_r1_k5_s2_e6_c32', 'ir_r1_k5_s1_e3_c32', 'ir_r1_k5_s1_e6_c32', 'ir_r1_k3_s1_e6_c32'],
['ir_r1_k5_s2_e6_c64', 'ir_r1_k5_s1_e3_c64', 'ir_r2_k5_s1_e6_c64'],
['ir_r3_k5_s1_e6_c112', 'ir_r1_k5_s1_e3_c112'],
['ir_r4_k5_s2_e6_c184'],
['ir_r1_k3_s1_e6_c352'],
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
stem_size=16,
num_features=1984, # paper suggests this, but is not 100% clear
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
**kwargs
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_spnasnet(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
Add Single-Path NAS pixel1 model
6 years ago
"""Creates the Single-Path NAS model from search targeted for Pixel1 phone.
Paper: https://arxiv.org/abs/1904.02877
Args:
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier: multiplier to number of channels per layer.
Add Single-Path NAS pixel1 model
6 years ago
"""
arch_def = [
# stage 0, 112x112 in
['ds_r1_k3_s1_c16_noskip'],
# stage 1, 112x112 in
['ir_r3_k3_s2_e3_c24'],
# stage 2, 56x56 in
['ir_r1_k5_s2_e6_c40', 'ir_r3_k3_s1_e3_c40'],
# stage 3, 28x28 in
['ir_r1_k5_s2_e6_c80', 'ir_r3_k3_s1_e3_c80'],
# stage 4, 14x14in
['ir_r1_k5_s1_e6_c96', 'ir_r3_k5_s1_e3_c96'],
# stage 5, 14x14in
['ir_r4_k5_s2_e6_c192'],
# stage 6, 7x7 in
['ir_r1_k3_s1_e6_c320_noskip']
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def),
Add Single-Path NAS pixel1 model
6 years ago
stem_size=32,
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Add Single-Path NAS pixel1 model
6 years ago
**kwargs
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Add Single-Path NAS pixel1 model
6 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_efficientnet(variant, channel_multiplier=1.0, depth_multiplier=1.0, pretrained=False, **kwargs):
Add drop_connect impl to try during training, fix a few comments
6 years ago
"""Creates an EfficientNet model.
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
Ref impl: https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/efficientnet_model.py
Paper: https://arxiv.org/abs/1905.11946
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
EfficientNet params
name: (channel_multiplier, depth_multiplier, resolution, dropout_rate)
'efficientnet-b0': (1.0, 1.0, 224, 0.2),
'efficientnet-b1': (1.0, 1.1, 240, 0.2),
'efficientnet-b2': (1.1, 1.2, 260, 0.3),
'efficientnet-b3': (1.2, 1.4, 300, 0.3),
'efficientnet-b4': (1.4, 1.8, 380, 0.4),
'efficientnet-b5': (1.6, 2.2, 456, 0.4),
'efficientnet-b6': (1.8, 2.6, 528, 0.5),
'efficientnet-b7': (2.0, 3.1, 600, 0.5),
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
'efficientnet-b8': (2.2, 3.6, 672, 0.5),
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
'efficientnet-l2': (4.3, 5.3, 800, 0.5),
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
Args:
channel_multiplier: multiplier to number of channels per layer
depth_multiplier: multiplier to number of repeats per stage
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
"""
arch_def = [
['ds_r1_k3_s1_e1_c16_se0.25'],
['ir_r2_k3_s2_e6_c24_se0.25'],
['ir_r2_k5_s2_e6_c40_se0.25'],
['ir_r3_k3_s2_e6_c80_se0.25'],
['ir_r3_k5_s1_e6_c112_se0.25'],
['ir_r4_k5_s2_e6_c192_se0.25'],
['ir_r1_k3_s1_e6_c320_se0.25'],
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def, depth_multiplier),
num_features=round_channels(1280, channel_multiplier, 8, None),
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
stem_size=32,
channel_multiplier=channel_multiplier,
Monster commit, activation refactor, VoVNet, norm_act improvements, more * refactor activations into basic PyTorch, jit scripted, and memory efficient custom auto * implement hard-mish, better grad for hard-swish * add initial VovNet V1/V2 impl, fix #151 * VovNet and DenseNet first models to use NormAct layers (support BatchNormAct2d, EvoNorm, InplaceIABN) * Wrap IABN for any models that use it * make more models torchscript compatible (DPN, PNasNet, Res2Net, SelecSLS) and add tests
5 years ago
act_layer=resolve_act_layer(kwargs, 'swish'),
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
**kwargs,
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_efficientnet_edge(variant, channel_multiplier=1.0, depth_multiplier=1.0, pretrained=False, **kwargs):
Fixup a few comments, add PyTorch version aware Flatten and finish as_sequential for GenEfficientNet
5 years ago
""" Creates an EfficientNet-EdgeTPU model
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/edgetpu
"""
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
arch_def = [
# NOTE `fc` is present to override a mismatch between stem channels and in chs not
# present in other models
['er_r1_k3_s1_e4_c24_fc24_noskip'],
['er_r2_k3_s2_e8_c32'],
['er_r4_k3_s2_e8_c48'],
['ir_r5_k5_s2_e8_c96'],
['ir_r4_k5_s1_e8_c144'],
['ir_r2_k5_s2_e8_c192'],
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def, depth_multiplier),
num_features=round_channels(1280, channel_multiplier, 8, None),
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
stem_size=32,
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Monster commit, activation refactor, VoVNet, norm_act improvements, more * refactor activations into basic PyTorch, jit scripted, and memory efficient custom auto * implement hard-mish, better grad for hard-swish * add initial VovNet V1/V2 impl, fix #151 * VovNet and DenseNet first models to use NormAct layers (support BatchNormAct2d, EvoNorm, InplaceIABN) * Wrap IABN for any models that use it * make more models torchscript compatible (DPN, PNasNet, Res2Net, SelecSLS) and add tests
5 years ago
act_layer=resolve_act_layer(kwargs, 'relu'),
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
**kwargs,
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
return model
def _gen_efficientnet_condconv(
variant, channel_multiplier=1.0, depth_multiplier=1.0, experts_multiplier=1, pretrained=False, **kwargs):
Fixup a few comments, add PyTorch version aware Flatten and finish as_sequential for GenEfficientNet
5 years ago
"""Creates an EfficientNet-CondConv model.
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
Fixup a few comments, add PyTorch version aware Flatten and finish as_sequential for GenEfficientNet
5 years ago
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/condconv
"""
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
arch_def = [
Replace all None by nn.Identity() in all models reset_classifier when False-values num_classes is given. Make small code refactoring
5 years ago
['ds_r1_k3_s1_e1_c16_se0.25'],
['ir_r2_k3_s2_e6_c24_se0.25'],
['ir_r2_k5_s2_e6_c40_se0.25'],
['ir_r3_k3_s2_e6_c80_se0.25'],
['ir_r3_k5_s1_e6_c112_se0.25_cc4'],
['ir_r4_k5_s2_e6_c192_se0.25_cc4'],
['ir_r1_k3_s1_e6_c320_se0.25_cc4'],
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
]
Fixup a few comments, add PyTorch version aware Flatten and finish as_sequential for GenEfficientNet
5 years ago
# NOTE unlike official impl, this one uses `cc<x>` option where x is the base number of experts for each stage and
# the expert_multiplier increases that on a per-model basis as with depth/channel multipliers
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def, depth_multiplier, experts_multiplier=experts_multiplier),
num_features=round_channels(1280, channel_multiplier, 8, None),
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
stem_size=32,
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Monster commit, activation refactor, VoVNet, norm_act improvements, more * refactor activations into basic PyTorch, jit scripted, and memory efficient custom auto * implement hard-mish, better grad for hard-swish * add initial VovNet V1/V2 impl, fix #151 * VovNet and DenseNet first models to use NormAct layers (support BatchNormAct2d, EvoNorm, InplaceIABN) * Wrap IABN for any models that use it * make more models torchscript compatible (DPN, PNasNet, Res2Net, SelecSLS) and add tests
5 years ago
act_layer=resolve_act_layer(kwargs, 'swish'),
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
**kwargs,
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
return model
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
def _gen_efficientnet_lite(variant, channel_multiplier=1.0, depth_multiplier=1.0, pretrained=False, **kwargs):
"""Creates an EfficientNet-Lite model.
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/lite
Paper: https://arxiv.org/abs/1905.11946
EfficientNet params
name: (channel_multiplier, depth_multiplier, resolution, dropout_rate)
'efficientnet-lite0': (1.0, 1.0, 224, 0.2),
'efficientnet-lite1': (1.0, 1.1, 240, 0.2),
'efficientnet-lite2': (1.1, 1.2, 260, 0.3),
'efficientnet-lite3': (1.2, 1.4, 280, 0.3),
'efficientnet-lite4': (1.4, 1.8, 300, 0.3),
Args:
channel_multiplier: multiplier to number of channels per layer
depth_multiplier: multiplier to number of repeats per stage
"""
arch_def = [
['ds_r1_k3_s1_e1_c16'],
['ir_r2_k3_s2_e6_c24'],
['ir_r2_k5_s2_e6_c40'],
['ir_r3_k3_s2_e6_c80'],
['ir_r3_k5_s1_e6_c112'],
['ir_r4_k5_s2_e6_c192'],
['ir_r1_k3_s1_e6_c320'],
]
model_kwargs = dict(
block_args=decode_arch_def(arch_def, depth_multiplier, fix_first_last=True),
num_features=1280,
stem_size=32,
fix_stem=True,
channel_multiplier=channel_multiplier,
Monster commit, activation refactor, VoVNet, norm_act improvements, more * refactor activations into basic PyTorch, jit scripted, and memory efficient custom auto * implement hard-mish, better grad for hard-swish * add initial VovNet V1/V2 impl, fix #151 * VovNet and DenseNet first models to use NormAct layers (support BatchNormAct2d, EvoNorm, InplaceIABN) * Wrap IABN for any models that use it * make more models torchscript compatible (DPN, PNasNet, Res2Net, SelecSLS) and add tests
5 years ago
act_layer=resolve_act_layer(kwargs, 'relu6'),
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
**kwargs,
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
return model
EfficientNet-V2S preliminary model def (for experimentation)
4 years ago
def _gen_efficientnet_v2s(variant, channel_multiplier=1.0, depth_multiplier=1.0, pretrained=False, **kwargs):
""" Creates an EfficientNet-V2s model
NOTE: this is a preliminary definition based on paper, awaiting official code release for details
and weights
Ref impl:
Update some comments re preliminary EfficientNet-V2 assumptions
4 years ago
Paper: `EfficientNetV2: Smaller Models and Faster Training` - https://arxiv.org/abs/2104.00298
EfficientNet-V2S preliminary model def (for experimentation)
4 years ago
"""
arch_def = [
Update some comments re preliminary EfficientNet-V2 assumptions
4 years ago
# FIXME it's not clear if the FusedMBConv layers have SE enabled for the Small variant,
# Table 4 suggests no. 23.94M params w/o, 23.98 with which is closer to 24M.
# ['er_r2_k3_s1_e1_c24_se0.25'],
# ['er_r4_k3_s2_e4_c48_se0.25'],
# ['er_r4_k3_s2_e4_c64_se0.25'],
['er_r2_k3_s1_e1_c24'],
EfficientNet-V2S preliminary model def (for experimentation)
4 years ago
['er_r4_k3_s2_e4_c48'],
['er_r4_k3_s2_e4_c64'],
['ir_r6_k3_s2_e4_c128_se0.25'],
['ir_r9_k3_s1_e6_c160_se0.25'],
['ir_r15_k3_s2_e6_c272_se0.25'],
]
model_kwargs = dict(
block_args=decode_arch_def(arch_def, depth_multiplier),
num_features=round_channels(1792, channel_multiplier, 8, None),
stem_size=24,
channel_multiplier=channel_multiplier,
norm_kwargs=resolve_bn_args(kwargs),
Update some comments re preliminary EfficientNet-V2 assumptions
4 years ago
act_layer=resolve_act_layer(kwargs, 'silu'), # FIXME this is an assumption, paper does not mention
EfficientNet-V2S preliminary model def (for experimentation)
4 years ago
**kwargs,
)
model = _create_effnet(variant, pretrained, **model_kwargs)
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_mixnet_s(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Small model.
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet/mixnet
Paper: https://arxiv.org/abs/1907.09595
"""
arch_def = [
# stage 0, 112x112 in
['ds_r1_k3_s1_e1_c16'], # relu
# stage 1, 112x112 in
['ir_r1_k3_a1.1_p1.1_s2_e6_c24', 'ir_r1_k3_a1.1_p1.1_s1_e3_c24'], # relu
# stage 2, 56x56 in
['ir_r1_k3.5.7_s2_e6_c40_se0.5_nsw', 'ir_r3_k3.5_a1.1_p1.1_s1_e6_c40_se0.5_nsw'], # swish
# stage 3, 28x28 in
['ir_r1_k3.5.7_p1.1_s2_e6_c80_se0.25_nsw', 'ir_r2_k3.5_p1.1_s1_e6_c80_se0.25_nsw'], # swish
# stage 4, 14x14in
['ir_r1_k3.5.7_a1.1_p1.1_s1_e6_c120_se0.5_nsw', 'ir_r2_k3.5.7.9_a1.1_p1.1_s1_e3_c120_se0.5_nsw'], # swish
# stage 5, 14x14in
['ir_r1_k3.5.7.9.11_s2_e6_c200_se0.5_nsw', 'ir_r2_k3.5.7.9_p1.1_s1_e6_c200_se0.5_nsw'], # swish
# 7x7
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def),
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
num_features=1536,
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
stem_size=16,
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
**kwargs
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def _gen_mixnet_m(variant, channel_multiplier=1.0, depth_multiplier=1.0, pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Medium-Large model.
Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet/mixnet
Paper: https://arxiv.org/abs/1907.09595
"""
arch_def = [
# stage 0, 112x112 in
['ds_r1_k3_s1_e1_c24'], # relu
# stage 1, 112x112 in
['ir_r1_k3.5.7_a1.1_p1.1_s2_e6_c32', 'ir_r1_k3_a1.1_p1.1_s1_e3_c32'], # relu
# stage 2, 56x56 in
['ir_r1_k3.5.7.9_s2_e6_c40_se0.5_nsw', 'ir_r3_k3.5_a1.1_p1.1_s1_e6_c40_se0.5_nsw'], # swish
# stage 3, 28x28 in
['ir_r1_k3.5.7_s2_e6_c80_se0.25_nsw', 'ir_r3_k3.5.7.9_a1.1_p1.1_s1_e6_c80_se0.25_nsw'], # swish
# stage 4, 14x14in
['ir_r1_k3_s1_e6_c120_se0.5_nsw', 'ir_r3_k3.5.7.9_a1.1_p1.1_s1_e3_c120_se0.5_nsw'], # swish
# stage 5, 14x14in
['ir_r1_k3.5.7.9_s2_e6_c200_se0.5_nsw', 'ir_r3_k3.5.7.9_p1.1_s1_e6_c200_se0.5_nsw'], # swish
# 7x7
]
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model_kwargs = dict(
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
block_args=decode_arch_def(arch_def, depth_multiplier, depth_trunc='round'),
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
num_features=1536,
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
stem_size=24,
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
channel_multiplier=channel_multiplier,
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
norm_kwargs=resolve_bn_args(kwargs),
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
**kwargs
)
Support for huggingface hub via create_model and default_cfgs. * improve consistency of model creation helper fns * add comments to some of the model helpers * support passing external default_cfgs so they can be sourced from hub
4 years ago
model = _create_effnet(variant, pretrained, **model_kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mnasnet_050(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet B1, depth multiplier of 0.5. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_b1('mnasnet_050', 0.5, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mnasnet_075(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet B1, depth multiplier of 0.75. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_b1('mnasnet_075', 0.75, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mnasnet_100(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet B1, depth multiplier of 1.0. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_b1('mnasnet_100', 1.0, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mnasnet_b1(pretrained=False, **kwargs):
Add the MnasNet-B1 variant weights, add A1/B1 model names as in the stand-alone repo, remove a bit of unused code
6 years ago
""" MNASNet B1, depth multiplier of 1.0. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
return mnasnet_100(pretrained, **kwargs)
Add the MnasNet-B1 variant weights, add A1/B1 model names as in the stand-alone repo, remove a bit of unused code
6 years ago
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mnasnet_140(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet B1, depth multiplier of 1.4 """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_b1('mnasnet_140', 1.4, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def semnasnet_050(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet A1 (w/ SE), depth multiplier of 0.5 """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_a1('semnasnet_050', 0.5, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def semnasnet_075(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet A1 (w/ SE), depth multiplier of 0.75. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_a1('semnasnet_075', 0.75, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def semnasnet_100(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet A1 (w/ SE), depth multiplier of 1.0. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_a1('semnasnet_100', 1.0, pretrained=pretrained, **kwargs)
Add support for tflite mnasnet pretrained weights and included spnasnet pretrained weights of my own. * tensorflow 'SAME' padding support added to GenMobileNet models for tflite pretrained weights * folded batch norm support (made batch norm optional and enable conv bias) for tflite pretrained weights * add url for spnasnet1_00 weights that I recently trained * fix SE reduction size for semnasnet models
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mnasnet_a1(pretrained=False, **kwargs):
Add the MnasNet-B1 variant weights, add A1/B1 model names as in the stand-alone repo, remove a bit of unused code
6 years ago
""" MNASNet A1 (w/ SE), depth multiplier of 1.0. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
return semnasnet_100(pretrained, **kwargs)
Add the MnasNet-B1 variant weights, add A1/B1 model names as in the stand-alone repo, remove a bit of unused code
6 years ago
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def semnasnet_140(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet A1 (w/ SE), depth multiplier of 1.4. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_a1('semnasnet_140', 1.4, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mnasnet_small(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" MNASNet Small, depth multiplier of 1.0. """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mnasnet_small('mnasnet_small', 1.0, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mobilenetv2_100(pretrained=False, **kwargs):
Add final weights for MobileNet-V2 experiments
5 years ago
""" MobileNet V2 w/ 1.0 channel multiplier """
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_mobilenet_v2('mobilenetv2_100', 1.0, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Experimenting with some MobileNetV2 variations to compare against EfficientNet-Lite
5 years ago
@register_model
Fiddling...
5 years ago
def mobilenetv2_140(pretrained=False, **kwargs):
Add final weights for MobileNet-V2 experiments
5 years ago
""" MobileNet V2 w/ 1.4 channel multiplier """
Fiddling...
5 years ago
model = _gen_mobilenet_v2('mobilenetv2_140', 1.4, pretrained=pretrained, **kwargs)
Experimenting with some MobileNetV2 variations to compare against EfficientNet-Lite
5 years ago
return model
@register_model
def mobilenetv2_110d(pretrained=False, **kwargs):
Add final weights for MobileNet-V2 experiments
5 years ago
""" MobileNet V2 w/ 1.1 channel, 1.2 depth multipliers"""
Fiddling...
5 years ago
model = _gen_mobilenet_v2(
Add final weights for MobileNet-V2 experiments
5 years ago
'mobilenetv2_110d', 1.1, depth_multiplier=1.2, fix_stem_head=True, pretrained=pretrained, **kwargs)
Experimenting with some MobileNetV2 variations to compare against EfficientNet-Lite
5 years ago
return model
@register_model
Fiddling...
5 years ago
def mobilenetv2_120d(pretrained=False, **kwargs):
Add final weights for MobileNet-V2 experiments
5 years ago
""" MobileNet V2 w/ 1.2 channel, 1.4 depth multipliers """
Fiddling...
5 years ago
model = _gen_mobilenet_v2(
Add final weights for MobileNet-V2 experiments
5 years ago
'mobilenetv2_120d', 1.2, depth_multiplier=1.4, fix_stem_head=True, pretrained=pretrained, **kwargs)
Experimenting with some MobileNetV2 variations to compare against EfficientNet-Lite
5 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def fbnetc_100(pretrained=False, **kwargs):
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
""" FBNet-C """
Add pretrained weights for FBNet-C
6 years ago
if pretrained:
# pretrained model trained with non-default BN epsilon
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_fbnetc('fbnetc_100', 1.0, pretrained=pretrained, **kwargs)
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related * add an alternate RMSprop opt that applies eps like TF * add bn params for passing through alternates and changing defaults to TF style
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def spnasnet_100(pretrained=False, **kwargs):
Add Single-Path NAS pixel1 model
6 years ago
""" Single-Path NAS Pixel1"""
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_spnasnet('spnasnet_100', 1.0, pretrained=pretrained, **kwargs)
Add Single-Path NAS pixel1 model
6 years ago
return model
Cleanup model_factory imports, consistent __all__ for models, fixed inception_v4 weight url
6 years ago
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b0(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B0 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b0', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b1(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B1 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b1', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b2(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B2 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.3, drop_path_rate should be 0.2
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b2', channel_multiplier=1.1, depth_multiplier=1.2, pretrained=pretrained, **kwargs)
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
return model
Add EfficientNet-B3 weights, trained from scratch with RA.
5 years ago
@register_model
def efficientnet_b2a(pretrained=False, **kwargs):
""" EfficientNet-B2 @ 288x288 w/ 1.0 test crop"""
Add updated EfficientNet-V2S weights, 83.8 @ 384x384 test. Add PyTorch trained EfficientNet-B4 weights, 83.4 @ 384x384 test. Tweak non TF EfficientNet B1-B4 train/test res scaling.
4 years ago
# WARN this model def is deprecated, different train/test res + test crop handled by default_cfg now
return efficientnet_b2(pretrained=pretrained, **kwargs)
Add EfficientNet-B3 weights, trained from scratch with RA.
5 years ago
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b3(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B3 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.3, drop_path_rate should be 0.2
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b3', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
return model
Add EfficientNet-B3 weights, trained from scratch with RA.
5 years ago
@register_model
def efficientnet_b3a(pretrained=False, **kwargs):
""" EfficientNet-B3 @ 320x320 w/ 1.0 test crop-pct """
Add updated EfficientNet-V2S weights, 83.8 @ 384x384 test. Add PyTorch trained EfficientNet-B4 weights, 83.4 @ 384x384 test. Tweak non TF EfficientNet B1-B4 train/test res scaling.
4 years ago
# WARN this model def is deprecated, different train/test res + test crop handled by default_cfg now
return efficientnet_b3(pretrained=pretrained, **kwargs)
Add EfficientNet-B3 weights, trained from scratch with RA.
5 years ago
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b4(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B4 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.4, drop_path_rate should be 0.2
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b4', channel_multiplier=1.4, depth_multiplier=1.8, pretrained=pretrained, **kwargs)
Add EfficientNet impl, change existing depth_multipler -> channel_multiplier as definitions have been confused
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b5(pretrained=False, **kwargs):
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
""" EfficientNet-B5 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.4, drop_path_rate should be 0.2
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b5', channel_multiplier=1.6, depth_multiplier=2.2, pretrained=pretrained, **kwargs)
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
return model
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b6(pretrained=False, **kwargs):
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
""" EfficientNet-B6 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.5, drop_path_rate should be 0.2
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b6', channel_multiplier=1.8, depth_multiplier=2.6, pretrained=pretrained, **kwargs)
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_b7(pretrained=False, **kwargs):
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
""" EfficientNet-B7 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.5, drop_path_rate should be 0.2
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_b7', channel_multiplier=2.0, depth_multiplier=3.1, pretrained=pretrained, **kwargs)
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
return model
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
@register_model
def efficientnet_b8(pretrained=False, **kwargs):
""" EfficientNet-B8 """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.5, drop_path_rate should be 0.2
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
model = _gen_efficientnet(
'efficientnet_b8', channel_multiplier=2.2, depth_multiplier=3.6, pretrained=pretrained, **kwargs)
return model
@register_model
def efficientnet_l2(pretrained=False, **kwargs):
""" EfficientNet-L2."""
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.5, drop_path_rate should be 0.2
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
model = _gen_efficientnet(
'efficientnet_l2', channel_multiplier=4.3, depth_multiplier=5.3, pretrained=pretrained, **kwargs)
return model
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_es(pretrained=False, **kwargs):
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
""" EfficientNet-Edge Small. """
model = _gen_efficientnet_edge(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_es', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
return model
adding efficientnet_el, efficientnet_es_prunedefficientnet_es_pruned and efficientnet_el_pruned pretrained models to timm/models
4 years ago
@register_model
def efficientnet_es_pruned(pretrained=False, **kwargs):
""" EfficientNet-Edge Small Pruned. For more info: https://github.com/DeGirum/pruned-models/releases/tag/efficientnet_v1.0"""
model = _gen_efficientnet_edge(
'efficientnet_es_pruned', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
return model
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_em(pretrained=False, **kwargs):
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
""" EfficientNet-Edge-Medium. """
model = _gen_efficientnet_edge(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_em', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_el(pretrained=False, **kwargs):
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
""" EfficientNet-Edge-Large. """
model = _gen_efficientnet_edge(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'efficientnet_el', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
return model
adding efficientnet_el, efficientnet_es_prunedefficientnet_es_pruned and efficientnet_el_pruned pretrained models to timm/models
4 years ago
@register_model
def efficientnet_el_pruned(pretrained=False, **kwargs):
""" EfficientNet-Edge-Large pruned. For more info: https://github.com/DeGirum/pruned-models/releases/tag/efficientnet_v1.0"""
model = _gen_efficientnet_edge(
'efficientnet_el_pruned', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
@register_model
def efficientnet_cc_b0_4e(pretrained=False, **kwargs):
""" EfficientNet-CondConv-B0 w/ 8 Experts """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_efficientnet_condconv(
'efficientnet_cc_b0_4e', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def efficientnet_cc_b0_8e(pretrained=False, **kwargs):
""" EfficientNet-CondConv-B0 w/ 8 Experts """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_efficientnet_condconv(
'efficientnet_cc_b0_8e', channel_multiplier=1.0, depth_multiplier=1.0, experts_multiplier=2,
pretrained=pretrained, **kwargs)
return model
Replace all None by nn.Identity() in all models reset_classifier when False-values num_classes is given. Make small code refactoring
5 years ago
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
@register_model
def efficientnet_cc_b1_8e(pretrained=False, **kwargs):
""" EfficientNet-CondConv-B1 w/ 8 Experts """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
model = _gen_efficientnet_condconv(
'efficientnet_cc_b1_8e', channel_multiplier=1.0, depth_multiplier=1.1, experts_multiplier=2,
pretrained=pretrained, **kwargs)
return model
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
@register_model
def efficientnet_lite0(pretrained=False, **kwargs):
""" EfficientNet-Lite0 """
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
model = _gen_efficientnet_lite(
'efficientnet_lite0', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
return model
@register_model
def efficientnet_lite1(pretrained=False, **kwargs):
""" EfficientNet-Lite1 """
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
model = _gen_efficientnet_lite(
'efficientnet_lite1', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
return model
@register_model
def efficientnet_lite2(pretrained=False, **kwargs):
""" EfficientNet-Lite2 """
# NOTE for train, drop_rate should be 0.3, drop_path_rate should be 0.2
model = _gen_efficientnet_lite(
'efficientnet_lite2', channel_multiplier=1.1, depth_multiplier=1.2, pretrained=pretrained, **kwargs)
return model
@register_model
def efficientnet_lite3(pretrained=False, **kwargs):
""" EfficientNet-Lite3 """
# NOTE for train, drop_rate should be 0.3, drop_path_rate should be 0.2
model = _gen_efficientnet_lite(
'efficientnet_lite3', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
return model
@register_model
def efficientnet_lite4(pretrained=False, **kwargs):
""" EfficientNet-Lite4 """
# NOTE for train, drop_rate should be 0.4, drop_path_rate should be 0.2
model = _gen_efficientnet_lite(
'efficientnet_lite4', channel_multiplier=1.4, depth_multiplier=1.8, pretrained=pretrained, **kwargs)
return model
added efficientnet pruned weights
5 years ago
@register_model
def efficientnet_b1_pruned(pretrained=False, **kwargs):
""" EfficientNet-B1 Pruned. The pruning has been obtained using https://arxiv.org/pdf/2002.08258.pdf """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
variant = 'efficientnet_b1_pruned'
model = _gen_efficientnet(
Move FeatureHooks into features.py, switch EfficientNet, MobileNetV3 to use build model helper
4 years ago
variant, channel_multiplier=1.0, depth_multiplier=1.1, pruned=True, pretrained=pretrained, **kwargs)
added efficientnet pruned weights
5 years ago
return model
@register_model
def efficientnet_b2_pruned(pretrained=False, **kwargs):
""" EfficientNet-B2 Pruned. The pruning has been obtained using https://arxiv.org/pdf/2002.08258.pdf """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
Move FeatureHooks into features.py, switch EfficientNet, MobileNetV3 to use build model helper
4 years ago
'efficientnet_b2_pruned', channel_multiplier=1.1, depth_multiplier=1.2, pruned=True,
pretrained=pretrained, **kwargs)
added efficientnet pruned weights
5 years ago
return model
@register_model
def efficientnet_b3_pruned(pretrained=False, **kwargs):
""" EfficientNet-B3 Pruned. The pruning has been obtained using https://arxiv.org/pdf/2002.08258.pdf """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
Move FeatureHooks into features.py, switch EfficientNet, MobileNetV3 to use build model helper
4 years ago
'efficientnet_b3_pruned', channel_multiplier=1.2, depth_multiplier=1.4, pruned=True,
pretrained=pretrained, **kwargs)
added efficientnet pruned weights
5 years ago
return model
EfficientNet-V2S preliminary model def (for experimentation)
4 years ago
@register_model
def efficientnet_v2s(pretrained=False, **kwargs):
""" EfficientNet-V2 Small. """
model = _gen_efficientnet_v2s(
'efficientnet_v2s', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
return model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
@register_model
def tf_efficientnet_b0(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B0. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b0', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_b1(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B1. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b1', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_b2(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B2. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b2', channel_multiplier=1.1, depth_multiplier=1.2, pretrained=pretrained, **kwargs)
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Update EfficientNet comments, MobileNetV3 non-TF create fns, fix factory arg checks, bump PyTorch version req to 1.2
5 years ago
def tf_efficientnet_b3(pretrained=False, **kwargs):
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
""" EfficientNet-B3. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b3', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
Ported Tensorflow pretrained EfficientNet weights and some model cleanup * B0-B3 weights ported from TF with close to paper accuracy * Renamed gen_mobilenet to gen_efficientnet since scaling params go well beyond 'mobile' specific * Add Tensorflow preprocessing option for closer images to source repo
6 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_b4(pretrained=False, **kwargs):
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
""" EfficientNet-B4. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b4', channel_multiplier=1.4, depth_multiplier=1.8, pretrained=pretrained, **kwargs)
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
return model
Add model registry and model listing fns, refactor model_factory/create_model fn
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_b5(pretrained=False, **kwargs):
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
""" EfficientNet-B5. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b5', channel_multiplier=1.6, depth_multiplier=2.2, pretrained=pretrained, **kwargs)
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_b6(pretrained=False, **kwargs):
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
""" EfficientNet-B6. Tensorflow compatible variant """
# NOTE for train, drop_rate should be 0.5
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b6', channel_multiplier=1.8, depth_multiplier=2.6, pretrained=pretrained, **kwargs)
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_b7(pretrained=False, **kwargs):
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
""" EfficientNet-B7. Tensorflow compatible variant """
# NOTE for train, drop_rate should be 0.5
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add new auto-augmentation Tensorflow EfficientNet weights, incl B6 and B7 models. Validation scores still pending but looking good.
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_b7', channel_multiplier=2.0, depth_multiplier=3.1, pretrained=pretrained, **kwargs)
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
return model
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
@register_model
def tf_efficientnet_b8(pretrained=False, **kwargs):
""" EfficientNet-B8. Tensorflow compatible variant """
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b8', channel_multiplier=2.2, depth_multiplier=3.6, pretrained=pretrained, **kwargs)
return model
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
@register_model
def tf_efficientnet_b0_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B0 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b0_ap', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b1_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B1 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b1_ap', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b2_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B2 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b2_ap', channel_multiplier=1.1, depth_multiplier=1.2, pretrained=pretrained, **kwargs)
return model
@register_model
Update EfficientNet comments, MobileNetV3 non-TF create fns, fix factory arg checks, bump PyTorch version req to 1.2
5 years ago
def tf_efficientnet_b3_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B3 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b3_ap', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b4_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B4 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b4_ap', channel_multiplier=1.4, depth_multiplier=1.8, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b5_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B5 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b5_ap', channel_multiplier=1.6, depth_multiplier=2.2, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b6_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B6 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b6_ap', channel_multiplier=1.8, depth_multiplier=2.6, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b7_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B7 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b7_ap', channel_multiplier=2.0, depth_multiplier=3.1, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b8_ap(pretrained=False, **kwargs):
Add newly added TF ported EfficientNet-B8 weights (RandAugment)
5 years ago
""" EfficientNet-B8 AdvProp. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b8_ap', channel_multiplier=2.2, depth_multiplier=3.6, pretrained=pretrained, **kwargs)
return model
Add ported EfficientNet-L2, B0-B7 NoisyStudent weights from TF TPU
5 years ago
@register_model
def tf_efficientnet_b0_ns(pretrained=False, **kwargs):
""" EfficientNet-B0 NoisyStudent. Tensorflow compatible variant """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b0_ns', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b1_ns(pretrained=False, **kwargs):
""" EfficientNet-B1 NoisyStudent. Tensorflow compatible variant """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b1_ns', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b2_ns(pretrained=False, **kwargs):
""" EfficientNet-B2 NoisyStudent. Tensorflow compatible variant """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b2_ns', channel_multiplier=1.1, depth_multiplier=1.2, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b3_ns(pretrained=False, **kwargs):
""" EfficientNet-B3 NoisyStudent. Tensorflow compatible variant """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b3_ns', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b4_ns(pretrained=False, **kwargs):
""" EfficientNet-B4 NoisyStudent. Tensorflow compatible variant """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b4_ns', channel_multiplier=1.4, depth_multiplier=1.8, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b5_ns(pretrained=False, **kwargs):
""" EfficientNet-B5 NoisyStudent. Tensorflow compatible variant """
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b5_ns', channel_multiplier=1.6, depth_multiplier=2.2, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b6_ns(pretrained=False, **kwargs):
""" EfficientNet-B6 NoisyStudent. Tensorflow compatible variant """
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b6_ns', channel_multiplier=1.8, depth_multiplier=2.6, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_b7_ns(pretrained=False, **kwargs):
""" EfficientNet-B7 NoisyStudent. Tensorflow compatible variant """
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_b7_ns', channel_multiplier=2.0, depth_multiplier=3.1, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_l2_ns_475(pretrained=False, **kwargs):
""" EfficientNet-L2 NoisyStudent @ 475x475. Tensorflow compatible variant """
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_l2_ns_475', channel_multiplier=4.3, depth_multiplier=5.3, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_l2_ns(pretrained=False, **kwargs):
""" EfficientNet-L2 NoisyStudent. Tensorflow compatible variant """
# NOTE for train, drop_rate should be 0.5
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet(
'tf_efficientnet_l2_ns', channel_multiplier=4.3, depth_multiplier=5.3, pretrained=pretrained, **kwargs)
return model
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_es(pretrained=False, **kwargs):
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
""" EfficientNet-Edge Small. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_edge(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_es', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_em(pretrained=False, **kwargs):
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
""" EfficientNet-Edge-Medium. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_edge(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_em', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_efficientnet_el(pretrained=False, **kwargs):
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
""" EfficientNet-Edge-Large. Tensorflow compatible variant """
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Finally got around to adding EdgeTPU EfficientNet variant
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_edge(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_efficientnet_el', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_cc_b0_4e(pretrained=False, **kwargs):
Fixup a few comments, add PyTorch version aware Flatten and finish as_sequential for GenEfficientNet
5 years ago
""" EfficientNet-CondConv-B0 w/ 4 Experts. Tensorflow compatible variant """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_condconv(
'tf_efficientnet_cc_b0_4e', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_cc_b0_8e(pretrained=False, **kwargs):
Fixup a few comments, add PyTorch version aware Flatten and finish as_sequential for GenEfficientNet
5 years ago
""" EfficientNet-CondConv-B0 w/ 8 Experts. Tensorflow compatible variant """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_condconv(
'tf_efficientnet_cc_b0_8e', channel_multiplier=1.0, depth_multiplier=1.0, experts_multiplier=2,
pretrained=pretrained, **kwargs)
return model
Replace all None by nn.Identity() in all models reset_classifier when False-values num_classes is given. Make small code refactoring
5 years ago
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
@register_model
def tf_efficientnet_cc_b1_8e(pretrained=False, **kwargs):
Fixup a few comments, add PyTorch version aware Flatten and finish as_sequential for GenEfficientNet
5 years ago
""" EfficientNet-CondConv-B1 w/ 8 Experts. Tensorflow compatible variant """
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_condconv(
'tf_efficientnet_cc_b1_8e', channel_multiplier=1.0, depth_multiplier=1.1, experts_multiplier=2,
pretrained=pretrained, **kwargs)
Add ported Tensorflow EfficientNet B4/B5 weights
5 years ago
return model
EfficientNet-Lite model added w/ converted checkpoints, validation in progress...
5 years ago
@register_model
def tf_efficientnet_lite0(pretrained=False, **kwargs):
""" EfficientNet-Lite0 """
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_lite(
'tf_efficientnet_lite0', channel_multiplier=1.0, depth_multiplier=1.0, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_lite1(pretrained=False, **kwargs):
""" EfficientNet-Lite1 """
# NOTE for train, drop_rate should be 0.2, drop_path_rate should be 0.2
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_lite(
'tf_efficientnet_lite1', channel_multiplier=1.0, depth_multiplier=1.1, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_lite2(pretrained=False, **kwargs):
""" EfficientNet-Lite2 """
# NOTE for train, drop_rate should be 0.3, drop_path_rate should be 0.2
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_lite(
'tf_efficientnet_lite2', channel_multiplier=1.1, depth_multiplier=1.2, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_lite3(pretrained=False, **kwargs):
""" EfficientNet-Lite3 """
# NOTE for train, drop_rate should be 0.3, drop_path_rate should be 0.2
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_lite(
'tf_efficientnet_lite3', channel_multiplier=1.2, depth_multiplier=1.4, pretrained=pretrained, **kwargs)
return model
@register_model
def tf_efficientnet_lite4(pretrained=False, **kwargs):
""" EfficientNet-Lite4 """
# NOTE for train, drop_rate should be 0.4, drop_path_rate should be 0.2
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same'
model = _gen_efficientnet_lite(
'tf_efficientnet_lite4', channel_multiplier=1.4, depth_multiplier=1.8, pretrained=pretrained, **kwargs)
return model
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mixnet_s(pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Small model.
"""
model = _gen_mixnet_s(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'mixnet_s', channel_multiplier=1.0, pretrained=pretrained, **kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mixnet_m(pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Medium model.
"""
model = _gen_mixnet_m(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'mixnet_m', channel_multiplier=1.0, pretrained=pretrained, **kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mixnet_l(pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Large model.
"""
model = _gen_mixnet_m(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'mixnet_l', channel_multiplier=1.3, pretrained=pretrained, **kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mixnet_xl(pretrained=False, **kwargs):
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
"""Creates a MixNet Extra-Large model.
Add weights for my MixNet-XL creation, include README updates for EdgeTPU models
5 years ago
Not a paper spec, experimental def by RW w/ depth scaling.
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
"""
model = _gen_mixnet_m(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'mixnet_xl', channel_multiplier=1.6, depth_multiplier=1.2, pretrained=pretrained, **kwargs)
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def mixnet_xxl(pretrained=False, **kwargs):
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
"""Creates a MixNet Double Extra Large model.
Add weights for my MixNet-XL creation, include README updates for EdgeTPU models
5 years ago
Not a paper spec, experimental def by RW w/ depth scaling.
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
"""
model = _gen_mixnet_m(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'mixnet_xxl', channel_multiplier=2.4, depth_multiplier=1.3, pretrained=pretrained, **kwargs)
Experimenting with a custom MixNet-XL and MixNet-XXL definition
5 years ago
return model
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_mixnet_s(pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Small model. Tensorflow compatible variant
"""
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_mixnet_s(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_mixnet_s', channel_multiplier=1.0, pretrained=pretrained, **kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_mixnet_m(pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Medium model. Tensorflow compatible variant
"""
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_mixnet_m(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_mixnet_m', channel_multiplier=1.0, pretrained=pretrained, **kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model
@register_model
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
def tf_mixnet_l(pretrained=False, **kwargs):
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
"""Creates a MixNet Large model. Tensorflow compatible variant
"""
Bring EfficientNet and MobileNetV3 up to date with my gen-efficientnet repo * Split MobileNetV3 and EfficientNet model files and put builder and blocks in own files (getting too large) * Finalize CondConv EfficientNet variant * Add the AdvProp weights files and B8 EfficientNet model * Refine the feature extraction module for EfficientNet and MobileNetV3
5 years ago
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
kwargs['pad_type'] = 'same'
model = _gen_mixnet_m(
Add CondConv support for EfficientNet into WIP for GenEfficientNet Feature extraction setup
5 years ago
'tf_mixnet_l', channel_multiplier=1.3, pretrained=pretrained, **kwargs)
Add MixNet (https://arxiv.org/abs/1907.09595) with pretrained weights converted from Tensorflow impl * refactor 'same' convolution and add helper to use MixedConv2d when needed * improve performance of 'same' padding for cases that can be handled statically * add support for extra exp, pw, and dw kernel specs with grouping support to decoder/string defs for MixNet * shuffle some args for a bit more consistency, a little less clutter overall in gen_efficientnet.py
5 years ago
return model