Migrations
/
pytorch-image-models
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Commit Graph

  • 878bee1d5e Add patch8 vit model to FX exclusion filter Ross Wightman 2021-11-22 14:00:27 -0800
  • 9bb4c80d2a Update README.md, missed recent regnetz results Ross Wightman 2021-11-22 12:08:46 -0800
  • 79bf4f163f Add mention of ResNet50 weights w/ RSB recipe ingredients and lower train res for better @ 224 results (but worse res scaling beyond). Not changing default. Ross Wightman 2021-11-22 11:49:05 -0800
  • ce76a810c2 New FX test strategy, filter based on param count Ross Wightman 2021-11-22 11:48:40 -0800
  • 1e51c2d02e More FX test tweaks Ross Wightman 2021-11-22 09:46:43 -0800
  • f0507f6da6 Fix k_decay default arg != 1.0 in poly scheduler Ross Wightman 2021-11-22 09:37:36 -0800
  • 41559247e9 use_ml_decoder_head talrid 2021-11-22 17:50:39 +0200
  • 947e1df3ef Update README.md Ross Wightman 2021-11-21 21:27:08 -0800
  • 90448031ea Filter more large models from FX tests Ross Wightman 2021-11-21 21:26:44 -0800
  • 1f53db2ece Updated lamhalobotnet weights, 81.5 top-1 Ross Wightman 2021-11-21 19:49:51 -0800
  • 8dc269c303 Filter more models for FX tests Ross Wightman 2021-11-21 19:49:33 -0800
  • 15ef108eb4 Add better halo2botnet50ts weights, 82 top-1 @ 256 Ross Wightman 2021-11-21 14:09:12 -0800
  • 2482652027 Add nfnet_f2 to FX test exclusion Ross Wightman 2021-11-21 14:08:53 -0800
  • 809c7bb1ec Merge remote-tracking branch 'origin/master' into bits_and_tpu Ross Wightman 2021-11-20 20:29:29 -0800
  • 734b2244fe Add RegNetZ-D8 (83.5 @ 256, 84 @ 320) and RegNetZ-E8 (84.5 @ 256, 85 @ 320) weights. Update names of existing RegZ models to include group size. Ross Wightman 2021-11-20 15:52:04 -0800
  • 05092e2fbe Add more models to FX filter Ross Wightman 2021-11-20 15:51:48 -0800
  • 93cc08fdc5 Make evonorm variables 1d to match other PyTorch norm layers, will break weight compat for any existing use (likely minimal, easy to fix). Ross Wightman 2021-11-20 15:50:51 -0800
  • af607b75cc Prep a set of ResNetV2 models with GroupNorm, EvoNormB0, EvoNormS0 for BN free model experiments on TPU and IPU Ross Wightman 2021-11-19 17:37:00 -0800
  • 3819bef93e Add FX test exclusion since it uses more ram and barfs on GitHub actions. Will take a few iterations to include needed models :( Ross Wightman 2021-11-19 17:35:41 -0800
  • c976a410d9 Add ResNet-50 w/ GN (resnet50_gn) and SEBotNet-33-TS (sebotnet33ts_256) model defs and weights. Update halonet50ts weights w/ slightly better variant in1k val, more robust to test sets. Ross Wightman 2021-11-19 14:24:43 -0800
  • 9b3519545d Attempt to reduce memory footprint of FX tests for GitHub actions runs Ross Wightman 2021-11-19 14:24:12 -0800
  • bdd3dff0ca beit_large models killing GitHub actions test, filter out Ross Wightman 2021-11-19 08:39:48 -0800
  • 9d6aad44f8 Update tests to run Python 3.9, PyTorch 1.10, torchvision 0.11.1 Ross Wightman 2021-11-18 21:38:04 -0800
  • ee40b582bb Merge branch 'alexander-soare-fx-feature-extract-new' Ross Wightman 2021-11-18 21:26:56 -0800
  • f2006b2437 Cleanup qkv_bias cat in beit model so it can be traced Ross Wightman 2021-11-18 21:25:00 -0800
  • 1076a65df1 Minor post FX merge cleanup Ross Wightman 2021-11-18 19:47:07 -0800
  • 32c9937dec Merge branch 'fx-feature-extract-new' of https://github.com/alexander-soare/pytorch-image-models into alexander-soare-fx-feature-extract-new Ross Wightman 2021-11-18 16:31:29 -0800
  • 78b36bf46c Places365 doesn't exist in some still used torchvision version Ross Wightman 2021-11-18 14:59:51 -0800
  • 27060edcfe
         Update helpers.py #967 jim4399266 2021-11-16 16:06:39 +0800
  • 65d827c7a6 rename notrace registration and standardize trace_utils imports #800 Alexander Soare 2021-11-15 21:03:21 +0000
  • 9b2daf2a35 Add ResNeXt-50 weights 81.1 top-1 @ 224, 82 @ 288 with A1 'high aug' recipe Ross Wightman 2021-11-14 13:17:27 -0800
  • 0a1ff9744a Merge branch 'martinsbruveris-add-vit-b8' Ross Wightman 2021-11-14 13:07:16 -0800
  • 9b5d6dc7e2 Merge branch 'add-vit-b8' of https://github.com/martinsbruveris/pytorch-image-models into martinsbruveris-add-vit-b8 Ross Wightman 2021-11-14 12:54:28 -0800
  • cfa414cad2 Matching two bits_and_tpu changes for TFDs wrapper * change 'samples' -> 'examples' for tfds wrapper to match tfds naming * add class_to_idx for image classification datasets in tfds wrapper Ross Wightman 2021-11-14 12:52:19 -0800
  • 5220711d87 Added B/8 models to ViT. #968 Martins Bruveris 2021-11-14 11:01:48 +0000
  • 0262a0e8e1 fx ready for review Alexander Soare 2021-11-13 00:06:33 +0000
  • 871cef4198 version 0.5.1 Ross Wightman 2021-11-12 13:53:55 -0800
  • 4f338556d8 Fixes and improvements for metrics, tfds parser, loader / transform handling * add back ability to create transform with loader * change 'samples' -> 'examples' for tfds wrapper to match tfds naming * add support for specifying feature names for input and target in tfds wrapper * add class_to_idx for image classification datasets in tfds wrapper * add accumulate_type to avg meters and metrics to allow float32 or float64 accumulation control with lower prec data * minor cleanup, log output rate prev and avg Ross Wightman 2021-11-12 13:40:26 -0800
  • d2994016e9 Add try/except guards Alexander Soare 2021-11-12 21:16:53 +0000
  • b25ff96768 wip - pre-rebase Alexander Soare 2021-11-12 20:42:45 +0000
  • e051dce354 Make all models FX traceable Alexander Soare 2021-08-12 15:31:24 +0100
  • cf4561ca72 Add FX based FeatureGraphNet capability Alexander Soare 2021-08-12 15:31:02 +0100
  • 0149ec30d7 wip - attempting to rebase Alexander Soare 2021-11-07 14:54:42 +0000
  • 02c3a75a45 wip - make it possible to use fx graph in train and eval mode Alexander Soare 2021-08-28 17:54:22 +0100
  • a6c24b936b Tests to enforce all models FX traceable Alexander Soare 2021-08-12 15:31:36 +0100
  • bc3d4eb403 wip -rebase Alexander Soare 2021-11-07 15:04:19 +0000
  • ab3ac3f25b Add FX based FeatureGraphNet capability Alexander Soare 2021-08-12 15:31:02 +0100
  • e16f526124
         Update efficientnet.py jim4399266 2021-11-12 11:06:49 +0800
  • 7ebb99e5d6
         Update efficientnet.py jim4399266 2021-11-12 11:05:33 +0800
  • caf28b6bc5
         Update helpers.py jim4399266 2021-11-12 10:49:56 +0800
  • a457b6d14d
         Update efficientnet.py jim4399266 2021-11-12 10:47:55 +0800
  • d9b0b3d60f device arg wasn't removed from PrefetcherCuda instantiation of RE Ross Wightman 2021-11-11 16:21:13 -0800
  • 80ca078aed Fix a few bugs and formatting/naming issues * Pass optimizer resume flag through to checkpoint / updater restore. Related to #961 but not clear how relates to crash. * Rename monitor step args, cleanup handling of step_end_idx vs num_steps for consistent log output in either case * Resume from proper epoch (ie next epoch relative to checkpoint) Ross Wightman 2021-11-11 15:09:31 -0800
  • 65419f60cc
         Merge pull request #964 from rwightman/more_datasets Ross Wightman 2021-11-11 13:34:05 -0800
  • 406c486ba2 Merge remote-tracking branch 'origin/more_datasets' into bits_and_tpu Ross Wightman 2021-11-10 16:23:21 -0800
  • 07693f81b0 Validation fix since we don't have multi-GPU DataParallel support yet Ross Wightman 2021-11-10 15:54:21 -0800
  • 9ec3210c2d More TFDS parser cleanup, support improved TFDS even_split impl (on tfds-nightly only currently). #964 more_datasets Ross Wightman 2021-11-10 15:52:09 -0800
  • ba65dfe2c6 Dataset work * support some torchvision datasets * improvements to TFDS wrapper for subsplit handling (fix #942), shuffle seed * add class-map support to train (fix #957) Ross Wightman 2021-11-09 22:34:15 -0800
  • 59a3409182
         Update README.md Ross Wightman 2021-11-03 17:10:41 -0700
  • ddc29da974 Add ResNet101 and ResNet152 weights from higher aug RSB recipes. 81.93 and 82.82 top-1 at 224x224. Ross Wightman 2021-11-02 17:58:14 -0700
  • b328e56f49 Update eca_halonext26ts weights to a better set Ross Wightman 2021-11-02 16:52:53 -0700
  • 2ddef942b9 Better fix for #954 that doesn't break torchscript, pull torch._assert into timm namespace when it exists Ross Wightman 2021-11-02 11:22:33 -0700
  • 4f0f9cb348 Fix #954 by bringing traceable _assert into timm to allow compat w/ PyTorch < 1.8 Ross Wightman 2021-11-02 09:21:40 -0700
  • 6cbc59628f add aa-layer arg for efficient-net class #603 Rahul Somani 2021-10-30 12:27:53 +0530
  • a41de1f666 Add interpolation mode handling to transforms. Removes InterpolationMode warning. Works for torchvision versions w/ and w/o InterpolationMode enum. Fix #738. Ross Wightman 2021-10-28 17:35:01 -0700
  • ed41d32637 Add repr to auto_augment and random_erasing impl Ross Wightman 2021-10-28 17:33:36 -0700
  • 135a48d024 Fix sam result again for imagenetv2 Ross Wightman 2021-10-28 14:46:45 -0700
  • aaff2d82d0 Add new 50ts attn models to benchmark/meta csv files Ross Wightman 2021-10-28 14:32:47 -0700
  • 1e17863b7b Fixed botne*t26 model results, add some 50ts self-attn variants Ross Wightman 2021-10-28 13:55:24 -0700
  • ae72d009fa Add weights for lambda_resnet50ts, halo2botnet50ts, lamhalobotnet50ts, updated halonet50ts Ross Wightman 2021-10-27 22:08:54 -0700
  • a45186a6e8 Merge remote-tracking branch 'origin/master' into bits_and_tpu Ross Wightman 2021-10-26 17:36:24 -0700
  • 13178ba73a Add benchmark and metadata csv files Ross Wightman 2021-10-26 10:50:46 -0700
  • b745d30a3e Fix formatting of last commit Ross Wightman 2021-10-25 15:15:14 -0700
  • 3478f1d7f1 Traceability fix for vit models for some experiments Ross Wightman 2021-10-25 15:13:08 -0700
  • f658a72e72 Cleanup re-use of Dropout modules in Mlp modules after some twitter feedback :p Ross Wightman 2021-10-25 00:40:59 -0700
  • 71f00bfe9e Don't run profile if model is torchscripted Ross Wightman 2021-10-24 22:41:20 -0700
  • 7da1b0b61c
         Merge pull request #933 from t-vi/unbind Ross Wightman 2021-10-24 22:33:18 -0700
  • 5882e62ada Add activation count to fvcore based profiling in benchmark.py Ross Wightman 2021-10-24 15:30:12 -0700
  • 51f488b7f5 Update results.csv files with latest weights Ross Wightman 2021-10-24 14:46:37 -0700
  • f805ba86d9 use .unbind instead of explicitly listing the indices #933 Thomas Viehmann 2021-10-24 21:08:47 +0200
  • 57992509f9 Fix some formatting in utils/model.py Ross Wightman 2021-10-23 20:35:36 -0700
  • 0fe4fd3f1f add d8 and e8 regnetz models with group size 8 Ross Wightman 2021-10-23 20:34:21 -0700
  • 690f31d02d Post merge cleanup, restore previous unwrap fn Ross Wightman 2021-10-23 20:30:53 -0700
  • 3b6ba76126 Merge remote-tracking branch 'origin/master' into bits_and_tpu Ross Wightman 2021-10-22 00:00:42 -0700
  • 25e7c8c5e5 Update broken resnetv2_50 weight url, add resnetv1_101 a1h recipe weights for 224x224 train Ross Wightman 2021-10-20 22:14:12 -0700
  • f7325c7b71 Support either deepspeed or fvcore for flop profiling Ross Wightman 2021-10-20 15:17:30 -0700
  • 66253790d4 Add `--bench profile` mode for benchmark.py to just run deepspeed detailed profile on model Ross Wightman 2021-10-19 16:06:38 -0700
  • 13a8bf7972 Add train size override and deepspeed GMACs counter (if deepspeed installed) to benchmark.py Ross Wightman 2021-10-19 15:14:26 -0700
  • 0ba73e6bcb
         Update README.md Ross Wightman 2021-10-19 14:38:56 -0700
  • b6caa356d2 Fixed eca_botnext26ts_256 weights added, 79.27 Ross Wightman 2021-10-19 12:44:28 -0700
  • c02334d9fa Add weights for regnetz_d and haloregnetz_c, update regnetz_c weights. Add commented PyTorch XLA code for halo attention Ross Wightman 2021-10-19 12:32:09 -0700
  • ad1c8f4f52
         Typo on the word "electricity" in opening statement #915 Jv Kyle Eclarin 2021-10-15 23:45:31 -0500
  • a21cceb7c4 Updated metrics.py script and used it in train.py #914 Abeer Alessa 2021-10-13 19:47:33 +0300
  • 02daf2ab94 Add option to include relative pos embedding in the attention scaling as per references. See discussion #912 Ross Wightman 2021-10-12 15:35:43 -0700
  • 2c33ca6d8c
         Merge pull request #913 from ground0state/master Ross Wightman 2021-10-12 14:09:56 -0700
  • 047a5ec05f Fix bugs that Mixup does not work device=cpu #913 masafumi 2021-10-12 23:51:46 +0900
  • cd34913278 Remove some outdated comments, botnet networks working great now. Ross Wightman 2021-10-11 22:43:41 -0700
  • 6ed4cdccca Update lambda_resnet26t weights with better set Ross Wightman 2021-10-10 16:32:54 -0700
  • 288ece0e9f
         Merge pull request #910 from tmp-iclr/master Ross Wightman 2021-10-10 16:00:58 -0700
  • 44d6d51668 Add ConvMixer #910 ICLR Author 2021-10-09 21:09:51 -0400