Fix MobileNetV2 head conv size for multiplier < 1.0. Add some missing modification copyrights, fix starting date of some old ones.

pull/1091/head
Ross Wightman 3 years ago committed by Ross Wightman
parent ccfeb06936
commit 656757d26b

@ -15,7 +15,7 @@ Papers:
RandAugment: Practical automated data augmentation... - https://arxiv.org/abs/1909.13719 RandAugment: Practical automated data augmentation... - https://arxiv.org/abs/1909.13719
AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty - https://arxiv.org/abs/1912.02781 AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty - https://arxiv.org/abs/1912.02781
Hacked together by / Copyright 2020 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import random import random
import math import math

@ -1,6 +1,6 @@
""" Quick n Simple Image Folder, Tarfile based DataSet """ Quick n Simple Image Folder, Tarfile based DataSet
Hacked together by / Copyright 2020 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import torch.utils.data as data import torch.utils.data as data
import os import os

@ -1,3 +1,7 @@
""" Dataset Factory
Hacked together by / Copyright 2021, Ross Wightman
"""
import os import os
from torchvision.datasets import CIFAR100, CIFAR10, MNIST, QMNIST, KMNIST, FashionMNIST, ImageNet, ImageFolder from torchvision.datasets import CIFAR100, CIFAR10, MNIST, QMNIST, KMNIST, FashionMNIST, ImageNet, ImageFolder

@ -3,7 +3,7 @@
Prefetcher and Fast Collate inspired by NVIDIA APEX example at Prefetcher and Fast Collate inspired by NVIDIA APEX example at
https://github.com/NVIDIA/apex/commit/d5e2bb4bdeedd27b1dfaf5bb2b24d6c000dee9be#diff-cf86c282ff7fba81fad27a559379d5bf https://github.com/NVIDIA/apex/commit/d5e2bb4bdeedd27b1dfaf5bb2b24d6c000dee9be#diff-cf86c282ff7fba81fad27a559379d5bf
Hacked together by / Copyright 2021 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import random import random
from functools import partial from functools import partial

@ -8,7 +8,7 @@ CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Fea
Code Reference: Code Reference:
CutMix: https://github.com/clovaai/CutMix-PyTorch CutMix: https://github.com/clovaai/CutMix-PyTorch
Hacked together by / Copyright 2020 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import numpy as np import numpy as np
import torch import torch

@ -3,7 +3,7 @@
Originally inspired by impl at https://github.com/zhunzhong07/Random-Erasing, Apache 2.0 Originally inspired by impl at https://github.com/zhunzhong07/Random-Erasing, Apache 2.0
Copyright Zhun Zhong & Liang Zheng Copyright Zhun Zhong & Liang Zheng
Hacked together by / Copyright 2020 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import random import random
import math import math

@ -1,7 +1,7 @@
""" Transforms Factory """ Transforms Factory
Factory methods for building image transforms for use with TIMM (PyTorch Image Models) Factory methods for building image transforms for use with TIMM (PyTorch Image Models)
Hacked together by / Copyright 2020 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import math import math

@ -4,6 +4,7 @@ Paper: 'Going deeper with Image Transformers' - https://arxiv.org/abs/2103.17239
Original code and weights from https://github.com/facebookresearch/deit, copyright below Original code and weights from https://github.com/facebookresearch/deit, copyright below
Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman
""" """
# Copyright (c) 2015-present, Facebook, Inc. # Copyright (c) 2015-present, Facebook, Inc.
# All rights reserved. # All rights reserved.

@ -9,6 +9,8 @@
Paper link: https://arxiv.org/abs/2103.10697 Paper link: https://arxiv.org/abs/2103.10697
Original code: https://github.com/facebookresearch/convit, original copyright below Original code: https://github.com/facebookresearch/convit, original copyright below
Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman
""" """
# Copyright (c) 2015-present, Facebook, Inc. # Copyright (c) 2015-present, Facebook, Inc.
# All rights reserved. # All rights reserved.

@ -12,6 +12,8 @@ Paper link: https://arxiv.org/abs/2103.14899
Original code: https://github.com/IBM/CrossViT/blob/main/models/crossvit.py Original code: https://github.com/IBM/CrossViT/blob/main/models/crossvit.py
NOTE: model names have been renamed from originals to represent actual input res all *_224 -> *_240 and *_384 -> *_408 NOTE: model names have been renamed from originals to represent actual input res all *_224 -> *_240 and *_384 -> *_408
Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman
""" """
# Copyright IBM All Rights Reserved. # Copyright IBM All Rights Reserved.

@ -409,7 +409,7 @@ class CspNet(nn.Module):
def _create_cspnet(variant, pretrained=False, **kwargs): def _create_cspnet(variant, pretrained=False, **kwargs):
cfg_variant = variant.split('_')[0] cfg_variant = variant.split('_')[0]
# NOTE: DarkNet is one of few models with stride==1 features w/ 6 out_indices [0..5] # NOTE: DarkNet is one of few models with stride==1 features w/ 6 out_indices [0..5]
out_indices = kwargs.get('out_indices', (0, 1, 2, 3, 4, 5) if 'darknet' in variant else (0, 1, 2, 3, 4)) out_indices = kwargs.pop('out_indices', (0, 1, 2, 3, 4, 5) if 'darknet' in variant else (0, 1, 2, 3, 4))
return build_model_with_cfg( return build_model_with_cfg(
CspNet, variant, pretrained, CspNet, variant, pretrained,
default_cfg=default_cfgs[variant], default_cfg=default_cfgs[variant],

@ -33,7 +33,7 @@ The majority of the above models (EfficientNet*, MixNet, MnasNet) and original w
by Mingxing Tan, Quoc Le, and other members of their Google Brain team. Thanks for consistently releasing by Mingxing Tan, Quoc Le, and other members of their Google Brain team. Thanks for consistently releasing
the models and weights open source! the models and weights open source!
Hacked together by / Copyright 2021 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
from functools import partial from functools import partial
from typing import List from typing import List
@ -718,7 +718,7 @@ def _gen_mobilenet_v2(
round_chs_fn = partial(round_channels, multiplier=channel_multiplier) round_chs_fn = partial(round_channels, multiplier=channel_multiplier)
model_kwargs = dict( model_kwargs = dict(
block_args=decode_arch_def(arch_def, depth_multiplier=depth_multiplier, fix_first_last=fix_stem_head), 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_chs_fn(1280), num_features=1280 if fix_stem_head else max(1280, round_chs_fn(1280)),
stem_size=32, stem_size=32,
fix_stem=fix_stem_head, fix_stem=fix_stem_head,
round_chs_fn=round_chs_fn, round_chs_fn=round_chs_fn,

@ -1,6 +1,6 @@
""" EfficientNet, MobileNetV3, etc Blocks """ EfficientNet, MobileNetV3, etc Blocks
Hacked together by / Copyright 2020 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import torch import torch

@ -3,7 +3,7 @@
Assembles EfficieNet and related network feature blocks from string definitions. Assembles EfficieNet and related network feature blocks from string definitions.
Handles stride, dilation calculations, and selects feature extraction points. Handles stride, dilation calculations, and selects feature extraction points.
Hacked together by / Copyright 2020 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
import logging import logging

@ -21,7 +21,7 @@ from .helpers import to_ntuple, to_2tuple, to_3tuple, to_4tuple, make_divisible
from .inplace_abn import InplaceAbn from .inplace_abn import InplaceAbn
from .linear import Linear from .linear import Linear
from .mixed_conv2d import MixedConv2d from .mixed_conv2d import MixedConv2d
from .mlp import Mlp, GluMlp, GatedMlp from .mlp import Mlp, GluMlp, GatedMlp, ConvMlp
from .non_local_attn import NonLocalAttn, BatNonLocalAttn from .non_local_attn import NonLocalAttn, BatNonLocalAttn
from .norm import GroupNorm, LayerNorm2d from .norm import GroupNorm, LayerNorm2d
from .norm_act import BatchNormAct2d, GroupNormAct from .norm_act import BatchNormAct2d, GroupNormAct

@ -14,7 +14,7 @@ Adapted from official impl at https://github.com/facebookresearch/LeViT, origina
This version combines both conv/linear models and fixes torchscript compatibility. This version combines both conv/linear models and fixes torchscript compatibility.
Modifications by/coyright Copyright 2021 Ross Wightman Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman
""" """
# Copyright (c) 2015-present, Facebook, Inc. # Copyright (c) 2015-present, Facebook, Inc.

@ -1,11 +1,10 @@
""" MobileNet V3 """ MobileNet V3
A PyTorch impl of MobileNet-V3, compatible with TF weights from official impl. A PyTorch impl of MobileNet-V3, compatible with TF weights from official impl.
Paper: Searching for MobileNetV3 - https://arxiv.org/abs/1905.02244 Paper: Searching for MobileNetV3 - https://arxiv.org/abs/1905.02244
Hacked together by / Copyright 2021 Ross Wightman Hacked together by / Copyright 2019, Ross Wightman
""" """
from functools import partial from functools import partial
from typing import List from typing import List

@ -4,7 +4,8 @@ This started as a copy of https://github.com/pytorch/vision 'resnet.py' (BSD-3-C
additional dropout and dynamic global avg/max pool. additional dropout and dynamic global avg/max pool.
ResNeXt, SE-ResNeXt, SENet, and MXNet Gluon stem/downsample variants, tiered stems added by Ross Wightman ResNeXt, SE-ResNeXt, SENet, and MXNet Gluon stem/downsample variants, tiered stems added by Ross Wightman
Copyright 2020 Ross Wightman
Copyright 2019, Ross Wightman
""" """
import math import math
from functools import partial from functools import partial

@ -4,6 +4,7 @@ A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shi
Code/weights from https://github.com/microsoft/Swin-Transformer, original copyright/license info below Code/weights from https://github.com/microsoft/Swin-Transformer, original copyright/license info below
Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman
""" """
# -------------------------------------------------------- # --------------------------------------------------------
# Swin Transformer # Swin Transformer

@ -180,7 +180,7 @@ def _filter_fn(state_dict):
def _create_vgg(variant: str, pretrained: bool, **kwargs: Any) -> VGG: def _create_vgg(variant: str, pretrained: bool, **kwargs: Any) -> VGG:
cfg = variant.split('_')[0] cfg = variant.split('_')[0]
# NOTE: VGG is one of few models with stride==1 features w/ 6 out_indices [0..5] # NOTE: VGG is one of few models with stride==1 features w/ 6 out_indices [0..5]
out_indices = kwargs.get('out_indices', (0, 1, 2, 3, 4, 5)) out_indices = kwargs.pop('out_indices', (0, 1, 2, 3, 4, 5))
model = build_model_with_cfg( model = build_model_with_cfg(
VGG, variant, pretrained, VGG, variant, pretrained,
default_cfg=default_cfgs[variant], default_cfg=default_cfgs[variant],

@ -4,6 +4,7 @@ Paper: Visformer: The Vision-friendly Transformer - https://arxiv.org/abs/2104.1
From original at https://github.com/danczs/Visformer From original at https://github.com/danczs/Visformer
Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman
""" """
from copy import deepcopy from copy import deepcopy

@ -20,7 +20,7 @@ for some einops/einsum fun
* Simple transformer style inspired by Andrej Karpathy's https://github.com/karpathy/minGPT * Simple transformer style inspired by Andrej Karpathy's https://github.com/karpathy/minGPT
* Bert reference code checks against Huggingface Transformers and Tensorflow Bert * Bert reference code checks against Huggingface Transformers and Tensorflow Bert
Hacked together by / Copyright 2021 Ross Wightman Hacked together by / Copyright 2020, Ross Wightman
""" """
import math import math
import logging import logging

@ -11,7 +11,7 @@ A PyTorch implement of the Hybrid Vision Transformers as described in:
NOTE These hybrid model definitions depend on code in vision_transformer.py. NOTE These hybrid model definitions depend on code in vision_transformer.py.
They were moved here to keep file sizes sane. They were moved here to keep file sizes sane.
Hacked together by / Copyright 2021 Ross Wightman Hacked together by / Copyright 2020, Ross Wightman
""" """
from copy import deepcopy from copy import deepcopy
from functools import partial from functools import partial

@ -1,10 +1,12 @@
""" Cross-Covariance Image Transformer (XCiT) in PyTorch """ Cross-Covariance Image Transformer (XCiT) in PyTorch
Same as the official implementation, with some minor adaptations.
- https://github.com/facebookresearch/xcit/blob/master/xcit.py
Paper: Paper:
- https://arxiv.org/abs/2106.09681 - https://arxiv.org/abs/2106.09681
Same as the official implementation, with some minor adaptations, original copyright below
- https://github.com/facebookresearch/xcit/blob/master/xcit.py
Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman
""" """
# Copyright (c) 2015-present, Facebook, Inc. # Copyright (c) 2015-present, Facebook, Inc.
# All rights reserved. # All rights reserved.

Loading…
Cancel
Save