Commit Graph

431 Commits (288682796f4a18760f1276dd2d184f297fdf5182)

Author SHA1 Message Date
Ross Wightman 0356e773f5 Default to native PyTorch AMP instead of APEX amp. Too many APEX issues cropping up lately.
4 years ago
Reuben 94ca140b67 update collections.abc import
4 years ago
Ross Wightman b4e216e377 Fix a few small things.
4 years ago
Ross Wightman dc85e5a237 Add ByobNet w/ GPU-EfficientNets and RepVGG. Also add classic vgg models.
4 years ago
Ross Wightman 1bcc69e0ad Use in_channels for depthwise groups, allows using `out_channels=N * in_channels` (does not impact existing models). Fix #354.
4 years ago
Ross Wightman 9811e229f7 Fix regression in models with 1001 class pretrained weights. Improve batchnorm arg and BatchNormAct layer handling in several models.
4 years ago
Ross Wightman a39c3ee216
Merge branch 'master' into eca-weights
4 years ago
Ross Wightman e9d6fe293c Update README for new weights. Version 0.4.2
4 years ago
Ross Wightman 666de85cf1 Move stride in EdgeResidual block to 3x3 expansion conv. Fix #414
4 years ago
Ross Wightman 3b57490a63 Fix some half removed resnet model defs, pooling for ecaresnet269d
4 years ago
Ross Wightman 68a4144882 Add new weights for ecaresnet26t/50t/269d models. Remove distinction between 't' and 'tn' (tiered models), tn is now t. Add test time img size spec to default cfg.
4 years ago
Ross Wightman b9843f954b
Merge pull request #282 from tigert1998/patch-1
4 years ago
hwangdeyu 7a4be5c035 add operator HardSwishJitAutoFn export to onnx
4 years ago
Ross Wightman 4203efa36d Fix #387 so that checkpoint saver works with max history of 1. Add checkpoint-hist arg to train.py.
4 years ago
Ross Wightman f0e65e37b7 Fix NF-ResNet101 model defs
4 years ago
Ross Wightman 2c988c3b6e Update README.md for NF-nets, bump version to 0.4.1 for merge
4 years ago
Ross Wightman 2de54d174a Fix pool size defs for NFNet models, add a comment.
4 years ago
Ross Wightman 90980de4a9 Fix up a few details in NFResNet models, managed stable training. Add support for gamma gain to be applied in activation or ScaleStdConv. Some tweaks to ScaledStdConv.
4 years ago
Ross Wightman 5a8e1e643e Initial Normalizer-Free Reg/ResNet impl. A bit of related layer refactoring.
4 years ago
Ross Wightman 38d8f67570 Fix potential issue with change to num_classes arg in train/validate.py defaulting to None (rely on model def / default_cfg)
4 years ago
Ross Wightman 587780e56b Update README.md and bump version to 0.4.0
4 years ago
Ross Wightman bb50ac4708 Add DeiT distilled weights and distilled model def. Remove some redudant ViT model args.
4 years ago
Ross Wightman c16e965037 Add some ViT comments and fix a few minor issues.
4 years ago
Ross Wightman 22748f1a2d Convert samples/targets in ParserImageInTar to numpy arrays, slightly less mem usage for massive datasets. Add a few more se/eca model defs to resnet.py
4 years ago
Ross Wightman 5d4c3d0af3 Add enhanced ParserImageInTar that can read images from tars within tars, folders with multiple tars, etc. Additional comment cleanup.
4 years ago
Ross Wightman 55f7dfa9ea Refactor vision_transformer entrpy fns, add pos embedding resize support for fine tuning, add some deit models for testing
4 years ago
Ross Wightman d55bcc0fee Finishing adding stochastic depth support to BiT ResNetV2 models
4 years ago
Ross Wightman 855d6cc217 More dataset work including factories and a tensorflow datasets (TFDS) wrapper
4 years ago
Ross Wightman 20516abc18 Fix some broken tests for ResNetV2 BiT models
4 years ago
Ross Wightman 59ec7e6a53 Merge branch 'master' into imagenet21k_datasets_more
4 years ago
Ross Wightman e7a9ddf982
Merge pull request #334 from kecsap/links
4 years ago
Csaba Kertesz 7cae7e7035 Follow links during dataset scanning
4 years ago
Ross Wightman c96e9f99a0 Update version to 0.3.3
4 years ago
Ross Wightman 4e2533db77 Add 320x320 model default cfgs for 101D and 152D ResNets. Add SEResNet-152D weights and 320x320 cfg.
4 years ago
Ross Wightman 0167f749d3 Remove some old __future__ imports
4 years ago
Ross Wightman e35e9760a6 More work on dataset / parser split and imagenet21k (tar) support
4 years ago
Ross Wightman ce69de70d3 Add 21k weight urls to vision_transformer. Cleanup feature_info for preact ResNetV2 (BiT) models
4 years ago
Ross Wightman 231d04e91a ResNetV2 pre-act and non-preact model, w/ BiT pretrained weights and support for ViT R50 model. Tweaks for in21k num_classes passing. More to do... tests failing.
4 years ago
Ross Wightman de6046e213 Initial commit for dataset / parser reorg to support additional datasets / types
4 years ago
Ross Wightman 392595c7eb Add pool_size to default cfgs for new models to prevent tests from failing. Add explicit 200D_320 model entrypoint for next benchmark run.
4 years ago
Ross Wightman b1f1228a41 Add ResNet101D, 152D, and 200D weights, remove meh 66d model
4 years ago
Jasha 7c56c718f3 Configure create_optimizer with args.opt_args
4 years ago
Ross Wightman 9a25fdf3ad
Merge pull request #297 from rwightman/ema_simplify
4 years ago
Tymoteusz Wiśniewski de15b43865 Fix a bug with accuracy retrieving from RealLabels
4 years ago
Ross Wightman cd72e66eff Bug in last mod for features_only default_cfg
4 years ago
Ross Wightman 867a0e5a04 Add default_cfg back to models wrapped in feature extraction module as per discussion in #294.
4 years ago
Ross Wightman 4ca52d73d8 Add separate set and update method to ModelEmaV2
4 years ago
Ross Wightman 2ed8f24715 A few more changes for 0.3.2 maint release. Linear layer change for mobilenetv3 and inception_v3, support no bias for linear wrapper.
4 years ago
Ross Wightman 6504a42832 Version 0.3.2
4 years ago
Ross Wightman 460eba7f24 Work around casting issue with combination of native torch AMP and torchscript for Linear layers
4 years ago
Ross Wightman 5f4b6076d8 Fix inplace arg compat for GELU and PreLU via activation factory
4 years ago
Ross Wightman fd962c4b4a Native SiLU (Swish) op doesn't export to ONNX
4 years ago
Ross Wightman 27bbc70d71 Add back old ModelEma and rename new one to ModelEmaV2 to avoid compat breaks in dependant code. Shuffle train script, add a few comments, remove DataParallel support, support experimental torchscript training.
4 years ago
tigertang 43f2500c26
Add symbolic for SwishJitAutoFn to support onnx
4 years ago
Ross Wightman 9214ca0716 Simplifying EMA...
4 years ago
Ross Wightman 53aeed3499 ver 0.3.1
4 years ago
Ross Wightman 30ab4a1494 Fix issue in optim factory with sgd / eps flag. Bump version to 0.3.1
4 years ago
Ross Wightman 741572dc9d Bump version to 0.3.0 for pending PyPi push
4 years ago
Ross Wightman b401952caf Add newly added vision transformer large/base 224x224 weights ported from JAX official repo
4 years ago
Ross Wightman 61200db0ab in_chans=1 working w/ pretrained weights for vision_transformer
4 years ago
Ross Wightman e90edce438 Support native silu activation (aka swish). An optimized ver is available in PyTorch 1.7.
4 years ago
Ross Wightman da6cd2cc1f Fix regression for pretrained classifier loading when using entrypt functions directly
4 years ago
Ross Wightman f591e90b0d Make sure num_features attr is present in vit models as with others
4 years ago
Ross Wightman 4a3df7842a Fix topn metric view regression on PyTorch 1.7
4 years ago
Ross Wightman f944242cb0 Fix #262, num_classes arg mixup. Make vision_transformers a bit closer to other models wrt get/reset classfier/forward_features. Fix torchscript for ViT.
4 years ago
Ross Wightman 736f209e7d Update vision transformers to be compatible with official code. Port official ViT weights from jax impl.
4 years ago
Ross Wightman 477a78ed81 Fix optimizer factory regressin for optimizers like sgd/momentum that don't have an eps arg
4 years ago
Ross Wightman 27a93e9de7 Improve test crop for ViT models. Small now 77.85, added base weights at 79.35 top-1.
4 years ago
Ross Wightman d4db9e7977 Add small vision transformer weights. 77.42 top-1.
4 years ago
talrid 27fadaa922 asymmetric_loss
4 years ago
Ross Wightman f31933cb37 Initial Vision Transformer impl w/ patch and hybrid variants. Refactor tuple helpers.
4 years ago
Ross Wightman a4d8fea61e Add model based wd skip support. Improve cross version compat of optimizer factory. Fix #247
4 years ago
Ross Wightman 80078c47bb Add Adafactor and Adahessian optimizers, cleanup optimizer arg passing, add gradient clipping support.
4 years ago
Ross Wightman fcb6258877 Add missing leaky_relu layer factory defn, update Apex/Native loss scaler interfaces to support unscaled grad clipping. Bump ver to 0.2.2 for pending release.
4 years ago
Ross Wightman e8e2d9cabf Add DropPath (stochastic depth) to ReXNet and VoVNet. RegNet DropPath impl tweak and dedupe se args.
4 years ago
Ross Wightman e8ca45854c More models in sotabench, more control over sotabench run, dataset filename extraction consistency
4 years ago
Ross Wightman 9c406532bd Add EfficientNet-EdgeTPU-M (efficientnet_em) model trained natively in PyTorch. More sotabench fiddling.
4 years ago
Ross Wightman c40384f5bd Add ResNet weights. 80.5 (top-1) ResNet-50-D, 77.1 ResNet-34-D, 72.7 ResNet-18-D.
4 years ago
Ross Wightman 47a7b3b5b1 More flexible mixup mode, add 'half' mode.
4 years ago
Ross Wightman 532e3b417d Reorg of utils into separate modules
4 years ago
Ross Wightman 33f8a1bf36 Updated README, add wide_resnet50_2 and seresnext50_32x4d weights
4 years ago
Ross Wightman 751b0bba98 Add global_pool (--gp) arg changes to allow passing 'fast' easily for train/validate to avoid channels_last issue with AdaptiveAvgPool
4 years ago
Ross Wightman 9c297ec67d Cleanup Apex vs native AMP scaler state save/load. Cleanup CheckpointSaver a bit.
4 years ago
Ross Wightman 80c9d9cc72 Add 'fast' global pool option, remove redundant SEModule from tresnet, normal one is now 'fast'
4 years ago
Ross Wightman 90a01f47d1 hrnet features_only pretrained weight loading issue. Fix #232.
4 years ago
Ross Wightman 110a7c4982 AdaptiveAvgPool2d -> mean((2,3)) for all SE/attn layers to avoid NaN with AMP + channels_last layout. See https://github.com/pytorch/pytorch/issues/43992
4 years ago
Ross Wightman c2cd1a332e Improve torch amp support and add channels_last support for train/validate scripts
4 years ago
Ross Wightman 470220b1f4 Fix MobileNetV3 crash with global_pool='', output consistent with other models but not equivalent due to efficient head.
4 years ago
Ross Wightman fc8b8afb6f Fix a silly bug in Sample version of EvoNorm missing x* part of swish, update EvoNormBatch to accumulated unbiased variance.
4 years ago
Ross Wightman f614df7921 Bump version to 0.2.1 and update README
4 years ago
Ross Wightman b423bc8362
Merge pull request #218 from rwightman/cutmix
4 years ago
Ross Wightman 8c9814e3f5 Final cleanup of mixup/cutmix. Element/batch modes working with both collate (prefetcher active) and without prefetcher.
4 years ago
Ross Wightman 0f5d9d8166 Add CSPResNet50 weights, 79.6 top-1 at 256x256
4 years ago
Ross Wightman b1b6e7c361 Fix a few more issues related to #216 w/ TResNet (space2depth) and FP16 weights in wide resnets. Also don't completely dump pretrained weights in in_chans != 1 or 3 cases.
4 years ago
Ross Wightman 512b2dd645 Add new EfficientNet-B3 and RegNetY-3.2GF weights, both just over 82 top-1
4 years ago
Ross Wightman 6890300877 Add DropPath (stochastic depth) to RegNet
4 years ago
Ross Wightman cd23f55397 Fix mixed prec issues with new mixup code
4 years ago
Yusuke Uchida f6b56602f9 fix test_model_default_cfgs
4 years ago
Ross Wightman f471c17c9d More cutmix/mixup overhaul, ready to kick-off some trials.
4 years ago
Ross Wightman d5145fa4d5 Change default_cfg names for senet to include the legacy and match model names
4 years ago