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

Commit Graph

  • 8f6d638887 Update README.md Ross Wightman 2022-03-23 16:16:26 -0700
  • 1618527098 Add layer scale and parallel blocks to vision_transformer Ross Wightman 2022-03-23 16:09:07 -0700
  • c42be74621 Add attrib / comments about Swin-S3 (AutoFormerV2) weights Ross Wightman 2022-03-23 16:07:09 -0700
  • 474ac906a2 Add 'head norm first' convnext_tiny_hnf weights Ross Wightman 2022-03-23 16:06:00 -0700
  • dc51334cdc Fix pruned adapt for EfficientNet models that are now using BatchNormAct layers Ross Wightman 2022-03-22 20:33:01 -0700
  • 22152ad840
         Cosmetic fix #1187 Michael Monashev 2022-03-22 09:54:16 +0300
  • 024fc4d9ab version 0.6.1 for master Ross Wightman 2022-03-21 22:03:13 -0700
  • e1e037ba52 Fix bad tuple typing fix that was on XLA branch bust missed on master merge Ross Wightman 2022-03-21 22:00:33 -0700
  • 341b464a5a Remove redundant noise attr from Plateau scheduler (use parent) Ross Wightman 2022-03-21 22:00:03 -0700
  • 7514439573
         Merge pull request #1014 from rwightman/norm_norm_norm Ross Wightman 2022-03-21 21:51:21 -0700
  • 749856cf25 Merge remote-tracking branch 'origin/norm_norm_norm' into bits_and_tpu Ross Wightman 2022-03-21 17:40:47 -0700
  • ff21fdb41d Update README.md ready for merge #1014 norm_norm_norm Ross Wightman 2022-03-21 16:38:36 -0700
  • fe457c1996 Update SwinTransformerV2Cr post-merge, update with grad checkpointing / grad matcher * weight compat break, activate norm3 for final block of final stage (equivalent to pre-head norm, but while still in BLC shape) * remove fold/unfold for TPU compat, add commented out roll code for TPU * add option for end of stage norm in all stages * allow weight_init to be selected between pytorch default inits and xavier / moco style vit variant Ross Wightman 2022-03-21 14:50:28 -0700
  • b049a5c5c6 Merge remote-tracking branch 'origin/master' into norm_norm_norm Ross Wightman 2022-03-21 13:41:43 -0700
  • 7cdd164d77 Fix #1184, scheduler noise bug during merge madness Ross Wightman 2022-03-21 13:35:45 -0700
  • 9440a50c95 Merge branch 'mrT23-master' Ross Wightman 2022-03-21 12:30:02 -0700
  • d98aa47d12 Revert ml-decoder changes to model factory and train script Ross Wightman 2022-03-21 12:29:02 -0700
  • 04db5833eb
         Merge pull request #986 from hankyul2/master Ross Wightman 2022-03-21 12:13:51 -0700
  • b20665d379
         Merge pull request #1007 from qwertyforce/patch-1 Ross Wightman 2022-03-21 12:12:58 -0700
  • d66b4c2355
         Merge pull request #801 from developer0hye/patch-1 Ross Wightman 2022-03-21 12:11:53 -0700
  • 7a0994f581
         Merge pull request #1150 from ChristophReich1996/master Ross Wightman 2022-03-21 11:56:57 -0700
  • d757fecaac
         Merge pull request #1112 from ayasyrev/sched_noise_dup_code Ross Wightman 2022-03-21 11:52:37 -0700
  • 61d3493f87 Fix hf-hub handling when hf-hub is config source Ross Wightman 2022-03-21 11:12:55 -0700
  • 5f47518f27 Fix pit implementation to be clsoer to deit/levit re distillation head handling Ross Wightman 2022-03-21 11:12:14 -0700
  • 0862e6ebae Fix correctness of some group matching regex (no impact on result), some formatting, missed forward_head for resnet Ross Wightman 2022-03-19 14:58:54 -0700
  • 94bcdebd73 Add latest weights trained on TPU-v3 VM instances Ross Wightman 2022-03-18 21:31:13 -0700
  • 95739b45d7 Fix partially removed alt_lable impl from TFDS variant of ImageNet22/12k Ross Wightman 2022-03-17 17:57:05 -0700
  • 5e1be34a60 Add ImageNet-22k/12k TFDS dataset defs Ross Wightman 2022-03-17 17:02:39 -0700
  • 59ffab537c Fix mistake in wds sample slicing Ross Wightman 2022-03-17 14:42:40 -0700
  • ef57561d51 Fix some TPU (XLA) issues with swin transformer v2 Ross Wightman 2022-03-16 14:55:36 -0700
  • ab16a358bb Add log and continue handler for WDS errors, fix args.num_gpu for validation script fallback Ross Wightman 2022-03-16 11:44:29 -0700
  • 7eeaf521a0 use gopen in wds to open info file in case it's at a url/gs location Ross Wightman 2022-03-11 20:42:39 -0800
  • 229ac6b8d8 Fix alternate label handling in WDS parser to skip invalid alt labels Ross Wightman 2022-03-11 19:16:04 -0800
  • a444d4b891 Add alternative label support to WDS for imagenet22k/12k split, add 21k/22k/12k indices filters to results/ Ross Wightman 2022-03-09 10:17:12 -0800
  • fa754db940
         Add files via upload kira7005 2022-03-08 01:09:03 -0600
  • 76e2b723db
         Create sample.md kira7005 2022-03-08 01:07:10 -0600
  • da2796ae82 Add webdataset (WDS) support, update TFDS to make some naming in parsers more similar. Fix workers=0 compatibility. Add ImageNet22k/12k synset defs. Ross Wightman 2022-03-07 21:13:20 -0800
  • 3fce010ca8 Merge remote-tracking branch 'origin/norm_norm_norm' into bits_and_tpu Ross Wightman 2022-02-28 17:07:56 -0800
  • 0557c8257d Fix bug introduced in non layer_decay weight_decay application. Remove debug print, fix arg desc. Ross Wightman 2022-02-28 17:06:32 -0800
  • 15cc9eae3e Fix Swin v2 tuple type hint Ross Wightman 2022-02-28 16:44:24 -0800
  • bb85b09d2a swin v2 fixup for latest changes on norm_norm_norm / bits_and_tpu branch Ross Wightman 2022-02-28 16:39:16 -0800
  • 10fa42b143 Merge branch 'ChristophReich1996-master' into bits_and_tpu Ross Wightman 2022-02-28 16:33:52 -0800
  • c639a86c67 Change TFDS default to full re-shuffle (init) each epoch (for now) Ross Wightman 2022-02-28 16:28:42 -0800
  • a16ea1e355 Merge remote-tracking branch 'origin/norm_norm_norm' into bits_and_tpu Ross Wightman 2022-02-28 16:28:27 -0800
  • 372ad5fa0d Significant model refactor and additions: * All models updated with revised foward_features / forward_head interface * Vision transformer and MLP based models consistently output sequence from forward_features (pooling or token selection considered part of 'head') * WIP param grouping interface to allow consistent grouping of parameters for layer-wise decay across all model types * Add gradient checkpointing support to a significant % of models, especially popular architectures * Formatting and interface consistency improvements across models * layer-wise LR decay impl part of optimizer factory w/ scale support in scheduler * Poolformer and Volo architectures added Ross Wightman 2022-02-28 13:56:23 -0800
  • 1420c118df Missed comitting outstanding changes to default_cfg keys and test exclusions for swin v2 #1150 Ross Wightman 2022-02-23 19:50:26 -0800
  • c6e4b7895a Swin V2 CR impl refactor. * reformat and change some naming so closer to existing timm vision transformers * remove typing that wasn't adding clarity (or causing torchscript issues) * support non-square windows * auto window size adjust from image size * post-norm + main-branch no Ross Wightman 2022-02-23 17:28:52 -0800
  • 67d140446b Fix bug in classification head Christoph Reich 2022-02-20 22:28:05 +0100
  • 29add820ac Refactor (back to relative imports) Christoph Reich 2022-02-20 00:46:48 +0100
  • 74a04e0016 Add parameter to change normalization type Christoph Reich 2022-02-20 00:46:00 +0100
  • 2a4f6c13dd Create model functions Christoph Reich 2022-02-20 00:40:22 +0100
  • 87b4d7a29a Add get and reset classifier method Christoph Reich 2022-02-19 22:47:02 +0100
  • ff5f6bcd6c Check input resolution Christoph Reich 2022-02-19 22:42:02 +0100
  • 81bf0b4033 Change parameter names to match Swin V1 Christoph Reich 2022-02-19 22:37:22 +0100
  • f227b88831 Add initials (CR) to model and file Christoph Reich 2022-02-19 22:14:38 +0100
  • 90dc74c450 Add code from https://github.com/ChristophReich1996/Swin-Transformer-V2 and change docstring style to match timm Christoph Reich 2022-02-19 22:12:11 +0100
  • 15bf46e9e3
         Merge 2ed06dd250 into 7c67d6aca9 #1063 Michael Monashev 2022-02-04 09:44:18 +0300
  • 7c67d6aca9
         Update README.md v0.1-tpu-weights 0.5.x Ross Wightman 2022-02-02 09:15:20 -0800
  • 2c3870e107 semobilevit_s for good measure Ross Wightman 2022-01-31 22:36:09 -0800
  • bcaeb91b03 Version to 0.6.0, possible interface incompatibilities vs 0.5.x Ross Wightman 2022-01-31 15:41:08 -0800
  • 58ba49c8ef Add MobileViT models (w/ ByobNet base). Close #1038. Ross Wightman 2022-01-31 15:38:32 -0800
  • fafece230b Allow changing base lr batch size from 256 via arg Ross Wightman 2022-01-27 17:35:50 -0800
  • 7148039f9f Tweak base lr log Ross Wightman 2022-01-27 17:29:49 -0800
  • f82fb6b608 Add base lr w/ linear and sqrt scaling to train script Ross Wightman 2022-01-27 17:24:47 -0800
  • 066e490605 Merge branch 'norm_norm_norm' into bits_and_tpu Ross Wightman 2022-01-27 16:47:45 -0800
  • 5f81d4de23 Move DeiT to own file, vit getting crowded. Working towards fixing #1029, make pooling interface for transformers and mlp closer to convnets. Still working through some details... Ross Wightman 2022-01-26 22:53:57 -0800
  • 629a0c1b8a fix typy hint noise_range_t #1113 ayasyrev 2022-01-26 16:08:33 +0300
  • 6734cf56ed cosine_lr docstring, type hints ayasyrev 2022-01-26 15:36:31 +0300
  • cf57695938 sched noise dup code remove #1112 ayasyrev 2022-01-26 11:53:08 +0300
  • 95cfc9b3e8 Merge remote-tracking branch 'origin/master' into norm_norm_norm Ross Wightman 2022-01-25 22:20:45 -0800
  • abc9ba2544 Transitioning default_cfg -> pretrained_cfg. Improving handling of pretrained_cfg source (HF-Hub, files, timm config, etc). Checkpoint handling tweaks. Ross Wightman 2022-01-25 21:54:13 -0800
  • 07379c6d5d Add vit_base2_patch32_256 for a model between base_patch16 and patch32 with a slightly larger img size and width v0.1-mvit-weights Ross Wightman 2022-01-24 14:45:59 -0800
  • cf4334391e Update benchmark and validate scripts to output results in JSON with a fixed delimiter for use in multi-process launcher Ross Wightman 2022-01-24 11:13:58 -0800
  • 1331c145a3 Add train benchmark results, adjust name scheme for inference and train benchmark files. Ross Wightman 2022-01-23 14:07:40 -0800
  • a517bf6a7a
         Merge pull request #1105 from kozistr/refactor/remove-condition Ross Wightman 2022-01-21 13:40:22 -0800
  • 56a6b38f76 refactor: remove if-condition #1105 kozistr 2022-01-21 14:19:11 +0900
  • 447677616f version 0.5.5 Ross Wightman 2022-01-20 21:18:30 -0800
  • 499c4749d7 Add update NCHW and NHWC inference benchmark numbers for current models. Flip name of 'sam' vit models in results files Ross Wightman 2022-01-20 10:40:04 -0800
  • 83b40c5a58 Last batch of small model weights (for now). mobilenetv3_small 050/075/100 and updated mnasnet_small with lambc/lamb optimizer. Ross Wightman 2022-01-19 10:01:18 -0800
  • 7f73252716
         Merge pull request #1094 from Mi-Peng/lars Ross Wightman 2022-01-19 08:39:49 -0800
  • cdcd0a92ca fix lars #1094 Mi-Peng 2022-01-19 17:49:43 +0800
  • 2d4b7e7080 Update results csvs for latest release Ross Wightman 2022-01-18 22:55:40 -0800
  • 1aa617cb3b Add AvgPool2d anti-aliasing support to ResNet arch (as per OpenAI CLIP models), add a few blur aa models as well Ross Wightman 2022-01-18 21:57:24 -0800
  • f0f9eccda8 Add --fuser arg to train/validate/benchmark scripts to select jit fuser type Ross Wightman 2022-01-17 13:54:25 -0800
  • 010b486590 Add Dino pretrained weights (no head) for vit models. Add support to tests and helpers for models w/ no classifier (num_classes=0 in pretrained cfg) Ross Wightman 2022-01-17 12:20:02 -0800
  • 34fb4a9fed Adding a qk_scale argument to Attention block #1091 S.Cao 2022-01-17 13:16:40 -0600
  • 738a9cd635 unbiased=False for torch.var_mean path of ConvNeXt LN. Fix #1090 Ross Wightman 2022-01-17 09:25:06 -0800
  • e0c4eec4b6 Default conv_mlp to False across the board for ConvNeXt, causing issues on more setups than it's improving right now... v0.5.4 Ross Wightman 2022-01-16 14:20:08 -0800
  • b669f4a588 Add ConvNeXt 22k->1k fine-tuned and 384 22k-1k fine-tuned weights after testing Ross Wightman 2022-01-15 15:43:52 -0800
  • 6dcbaf211a
         Update README.md Ross Wightman 2022-01-14 20:11:45 -0800
  • a8d103e18b Giant/gigantic vits snuck through in a test a broke GitHub test runner, add filter Ross Wightman 2022-01-14 17:23:35 -0800
  • ef72ad4177 Extra vit_huge model likely to cause test issue (non in21k variant), adding to filters Ross Wightman 2022-01-14 15:56:59 -0800
  • e967c72875 Update REAMDE.md. Sneak in g/G (giant / gigantic?) ViT defs from scaling paper Ross Wightman 2022-01-14 15:39:31 -0800
  • 9ca3437178 Add some more small model weights lcnet, mnas, mnv2 Ross Wightman 2022-01-14 13:13:10 -0800
  • fa6463c936 Version 0.5.4 Ross Wightman 2022-01-14 12:23:50 -0800
  • fa81164378 Fix stem width for really small mobilenetv3 arch defs Ross Wightman 2022-01-13 22:42:02 -0800
  • edd3d73695 Add missing dropout for head reset in ConvNeXt default head Ross Wightman 2022-01-13 22:41:23 -0800
  • b093dcb46d Some convnext cleanup, remove in place mul_ for gamma, breaking symbolic trace, cleanup head a bit... Ross Wightman 2022-01-13 21:10:32 -0800
  • 18934debc5 Add initial ConvNeXt impl (mods of official code) Ross Wightman 2022-01-12 23:05:41 -0800
  • 656757d26b Fix MobileNetV2 head conv size for multiplier < 1.0. Add some missing modification copyrights, fix starting date of some old ones. Ross Wightman 2022-01-12 16:18:18 -0800