Commit Graph

  • 38d8f67570 Fix potential issue with change to num_classes arg in train/validate.py defaulting to None (rely on model def / default_cfg) Ross Wightman 2021-01-25 11:53:34 -0800
  • 587780e56b Update README.md and bump version to 0.4.0 Ross Wightman 2021-01-25 11:22:11 -0800
  • bb50ac4708 Add DeiT distilled weights and distilled model def. Remove some redudant ViT model args. Ross Wightman 2021-01-25 11:05:23 -0800
  • d17adadabb fix: add triplet attention to model registry iyaja 2021-01-25 13:19:19 +0530
  • c16e965037 Add some ViT comments and fix a few minor issues. Ross Wightman 2021-01-24 23:18:35 -0800
  • 5cdebbf396 fix: triplet attention module name iyaja 2021-01-25 00:41:41 +0530
  • 06b86a3f7a feat: add triplet attention layer iyaja 2021-01-25 00:32:36 +0530
  • 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 Ross Wightman 2021-01-22 16:54:33 -0800
  • 5d4c3d0af3 Add enhanced ParserImageInTar that can read images from tars within tars, folders with multiple tars, etc. Additional comment cleanup. Ross Wightman 2021-01-22 10:52:04 -0800
  • 23dc3f3974
    Update vision_transformer.py #367 Zhiyuan Chen 2021-01-20 17:57:19 +0800
  • 55f7dfa9ea Refactor vision_transformer entrpy fns, add pos embedding resize support for fine tuning, add some deit models for testing Ross Wightman 2021-01-18 16:11:02 -0800
  • 28c0fa31fe
    Update vision_transformer.py Zhiyuan Chen 2021-01-18 18:13:01 +0800
  • ee3d8050c8
    Align block with original design Zhiyuan Chen 2021-01-18 18:12:01 +0800
  • 9d5d4b8df6 Fix silly train.py typo during dataset work Ross Wightman 2021-01-16 16:32:21 -0800
  • d55bcc0fee Finishing adding stochastic depth support to BiT ResNetV2 models Ross Wightman 2021-01-16 16:32:03 -0800
  • 0a1668f63e Update tests Ross Wightman 2021-01-16 12:37:10 -0800
  • 58ccf43150 Add BiT references and knowledge distill links to readme/docs Ross Wightman 2021-01-16 12:30:02 -0800
  • 855d6cc217 More dataset work including factories and a tensorflow datasets (TFDS) wrapper * Add parser/dataset factory methods for more flexible dataset & parser creation * Add dataset parser that wraps TFDS image classification datasets * Tweak num_classes handling bug for 21k models * Add initial deit models so they can be benchmarked in next csv results runs Ross Wightman 2021-01-15 17:26:20 -0800
  • 201b0046f7
    Include pre_logits in vit #352 Zhiyuan Chen 2021-01-09 05:14:28 +0800
  • f8463b8fa9 Version 0.3.4. Tweak setup.cfg and update setup.py metadata v0.3.4 Ross Wightman 2021-01-05 21:24:28 -0800
  • 20516abc18 Fix some broken tests for ResNetV2 BiT models Ross Wightman 2021-01-04 23:21:39 -0800
  • fd9061dbf7 Remove debug print from train.py Ross Wightman 2021-01-04 12:16:06 -0800
  • 59ec7e6a53 Merge branch 'master' into imagenet21k_datasets_more Ross Wightman 2021-01-04 12:11:05 -0800
  • fc3d9183e8
    Merge pull request #335 from kecsap/new_option Ross Wightman 2021-01-04 10:32:21 -0800
  • e7a9ddf982
    Merge pull request #334 from kecsap/links Ross Wightman 2021-01-04 10:30:58 -0800
  • 19816fe226 Add citation info Ross Wightman 2021-01-03 15:46:12 -0800
  • e42b140ade Add --input-size option to scripts to specify full input dimensions from command-line #335 Csaba Kertesz 2021-01-04 00:25:29 +0200
  • 7cae7e7035 Follow links during dataset scanning #334 Csaba Kertesz 2021-01-04 00:16:45 +0200
  • 1d01c2b68c
    Update README.md v0.3.3 Ross Wightman 2021-01-03 12:49:31 -0800
  • c96e9f99a0 Update version to 0.3.3 Ross Wightman 2021-01-03 12:43:44 -0800
  • a7d0a8b5b2 Update results csv files with latest models, incl 101D, 152D, 200D, SE152D ResNets and yet to be merged BiT and ViT-R50 models. Ross Wightman 2021-01-03 12:42:43 -0800
  • 4e2533db77 Add 320x320 model default cfgs for 101D and 152D ResNets. Add SEResNet-152D weights and 320x320 cfg. Ross Wightman 2021-01-03 11:25:07 -0800
  • 0167f749d3 Remove some old __future__ imports Ross Wightman 2021-01-03 11:24:16 -0800
  • 85bf4b8cd6 Add setup.cfg for conda / fastai integration Ross Wightman 2021-01-03 11:20:33 -0800
  • e553480b67 Add 21843 synset txt for google 21k models like BiT/ViT Ross Wightman 2020-12-28 14:39:29 -0800
  • e35e9760a6 More work on dataset / parser split and imagenet21k (tar) support Ross Wightman 2020-12-28 14:39:05 -0800
  • ce69de70d3 Add 21k weight urls to vision_transformer. Cleanup feature_info for preact ResNetV2 (BiT) models Ross Wightman 2020-12-28 14:37:06 -0800
  • 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. Ross Wightman 2020-12-24 12:03:31 -0800
  • de6046e213 Initial commit for dataset / parser reorg to support additional datasets / types Ross Wightman 2020-12-08 17:03:00 -0800
  • 16cd1f9fd2 Change the --img-size parameter into a int list to specify depth, height and width by a parameter #322 Csaba Kertesz 2020-12-28 22:52:53 +0200
  • 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. Ross Wightman 2020-12-18 21:28:47 -0800
  • 7a75b8d033
    Update README.md Ross Wightman 2020-12-18 17:27:57 -0800
  • b1f1228a41 Add ResNet101D, 152D, and 200D weights, remove meh 66d model Ross Wightman 2020-12-18 17:13:37 -0800
  • 7ec4ef66bb
    download pretrained models with progress #315 hiyyg 2020-12-18 14:58:10 +0800
  • 198f6ea0f3
    Merge pull request #302 from Jasha10/create_optimizer-opt_args Ross Wightman 2020-12-08 10:19:36 -0800
  • 7c56c718f3 Configure create_optimizer with args.opt_args #302 Jasha 2020-12-08 00:03:09 -0600
  • 51d74d91da Update README.md Ross Wightman 2020-12-07 09:59:50 -0800
  • 9a25fdf3ad
    Merge pull request #297 from rwightman/ema_simplify Ross Wightman 2020-12-05 11:42:45 -0800
  • 18a5a6abb2
    Merge ff1a657f57 into c9ebe86d03 #290 Roman Martyanov 2020-12-04 08:36:49 -0800
  • c9ebe86d03
    Merge pull request #300 from tmkkk/real-labels-fix Ross Wightman 2020-12-04 08:36:38 -0800
  • de15b43865 Fix a bug with accuracy retrieving from RealLabels #300 Tymoteusz Wiśniewski 2020-12-04 16:12:50 +0100
  • cd72e66eff Bug in last mod for features_only default_cfg #297 Ross Wightman 2020-12-03 12:33:01 -0800
  • 867a0e5a04 Add default_cfg back to models wrapped in feature extraction module as per discussion in #294. Ross Wightman 2020-12-03 10:24:35 -0800
  • 4ca52d73d8 Add separate set and update method to ModelEmaV2 Ross Wightman 2020-12-03 10:05:09 -0800
  • 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. Ross Wightman 2020-11-30 16:19:52 -0800
  • 6504a42832 Version 0.3.2 Ross Wightman 2020-11-30 13:39:08 -0800
  • 460eba7f24 Work around casting issue with combination of native torch AMP and torchscript for Linear layers Ross Wightman 2020-11-30 13:29:33 -0800
  • 5f4b6076d8 Fix inplace arg compat for GELU and PreLU via activation factory Ross Wightman 2020-11-30 13:27:40 -0800
  • fd962c4b4a Native SiLU (Swish) op doesn't export to ONNX Ross Wightman 2020-11-29 21:56:55 -0800
  • 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. Ross Wightman 2020-11-29 16:12:41 -0800
  • ff1a657f57 fixing config reader #290 romamartyanov 2020-11-29 22:05:46 +0300
  • c14fab4710 Added class for easy model evaluation on one image romamartyanov 2020-11-28 18:07:49 +0300
  • fa2e5c6f16 Added set_deterministic function romamartyanov 2020-11-28 18:04:19 +0300
  • 41e1f8d282 Moved all argparse configs to .yaml files romamartyanov 2020-11-28 18:01:43 +0300
  • 6f43aeb252
    Merge pull request #286 from s-rog/patch-1 Ross Wightman 2020-11-24 09:47:15 -0800
  • a7f6126b92
    Update README.md #286 Roger Shieh 2020-11-24 09:02:37 +0800
  • e265d4bc31
    relaxe size constraints in vit #284 Zhiyuan Chen 2020-11-21 05:55:30 +0800
  • 43f2500c26
    Add symbolic for SwishJitAutoFn to support onnx #282 tigertang 2020-11-18 14:36:12 +0800
  • 9214ca0716 Simplifying EMA... Ross Wightman 2020-11-16 12:51:52 -0800
  • 80cd31f21f
    Create config.yml Ross Wightman 2020-11-12 20:55:27 -0800
  • d940a53cd9 Update issue templates Ross Wightman 2020-11-10 12:16:40 -0800
  • 60c998af22 Update issue templates Ross Wightman 2020-11-10 11:39:52 -0800
  • 9d73911e62 Update issue templates Ross Wightman 2020-11-10 11:35:30 -0800
  • 2542283b28
    Merge pull request #268 from seemethere/patch-1 Ross Wightman 2020-11-02 11:46:21 -0800
  • ab9f1fc995
    Update torch -> 1.7.0, torchvision -> 0.8.1 #268 Eli Uriegas 2020-11-02 10:20:17 -0800
  • 53aeed3499 ver 0.3.1 Ross Wightman 2020-10-31 18:14:58 -0700
  • 30ab4a1494 Fix issue in optim factory with sgd / eps flag. Bump version to 0.3.1 Ross Wightman 2020-10-31 18:03:35 -0700
  • da8447c4ae Deployed 46f1544 with MkDocs version: 1.1.2 Ross Wightman 2020-10-30 12:31:19 -0800
  • 46f15443be Update README.md and docs in prep for 0.3.0 PyPi release. Ross Wightman 2020-10-30 13:27:19 -0700
  • af3299ba4a
    Merge pull request #263 from rwightman/fixes_oct2020 Ross Wightman 2020-10-29 20:14:13 -0700
  • 741572dc9d Bump version to 0.3.0 for pending PyPi push #263 Ross Wightman 2020-10-29 17:31:39 -0700
  • b401952caf Add newly added vision transformer large/base 224x224 weights ported from JAX official repo Ross Wightman 2020-10-29 17:31:01 -0700
  • 61200db0ab in_chans=1 working w/ pretrained weights for vision_transformer Ross Wightman 2020-10-29 15:49:36 -0700
  • e90edce438 Support native silu activation (aka swish). An optimized ver is available in PyTorch 1.7. Ross Wightman 2020-10-29 15:45:17 -0700
  • da6cd2cc1f Fix regression for pretrained classifier loading when using entrypt functions directly Ross Wightman 2020-10-29 15:43:39 -0700
  • f591e90b0d Make sure num_features attr is present in vit models as with others Ross Wightman 2020-10-29 15:33:47 -0700
  • 4a3df7842a Fix topn metric view regression on PyTorch 1.7 Ross Wightman 2020-10-29 14:04:15 -0700
  • 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. Ross Wightman 2020-10-29 13:58:28 -0700
  • da1b90e5c9 Update results csvs with latest run Ross Wightman 2020-10-29 07:55:13 -0700
  • 736f209e7d Update vision transformers to be compatible with official code. Port official ViT weights from jax impl. Ross Wightman 2020-10-26 18:42:11 -0700
  • 7613094fb5 Add ViT to sotabench v0.1-vitjx Ross Wightman 2020-10-22 16:07:22 -0700
  • 477a78ed81 Fix optimizer factory regressin for optimizers like sgd/momentum that don't have an eps arg Ross Wightman 2020-10-22 15:59:47 -0700
  • 27a93e9de7 Improve test crop for ViT models. Small now 77.85, added base weights at 79.35 top-1. Ross Wightman 2020-10-21 23:35:25 -0700
  • d4db9e7977 Add small vision transformer weights. 77.42 top-1. Ross Wightman 2020-10-21 12:14:12 -0700
  • f973722adb vision transform forward_features #256 Zvi Lapp 2020-10-18 18:24:44 +0300
  • ccfb5751ab
    Merge pull request #255 from mrT23/master Ross Wightman 2020-10-16 20:26:33 -0700
  • 27fadaa922 asymmetric_loss #255 talrid 2020-10-16 17:12:28 +0300
  • 79e727e07a Merge branch 'master' of https://github.com/mrT23/pytorch-image-models talrid 2020-10-16 16:55:14 +0300
  • 8331fac688
    Merge pull request #2 from rwightman/master mrT23 2020-10-16 16:55:00 +0300
  • 70ae7f0cc2
    Merge pull request #250 from rwightman/vision_transformer Ross Wightman 2020-10-13 15:29:37 -0700