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

Commit Graph

  • e939ed19b9 Rename internal creation fn for maxvit, has not been just coatnet for a while... Ross Wightman 2022-08-22 17:44:51 -0700
  • ffaf97f813 MaxxVit! A very configurable MaxVit and CoAtNet impl with lots of goodies.. Ross Wightman 2022-08-22 17:42:10 -0700
  • 8c9696c9df More model and test fixes Ross Wightman 2022-08-22 17:40:31 -0700
  • ca52108c2b Fix some model support functions Ross Wightman 2022-08-19 10:20:51 -0700
  • e710cc041f
         added training script note #1420 Florian 2022-08-19 14:33:36 +0200
  • f332fc2db7 Fix some test failures, torchscript issues Ross Wightman 2022-08-18 16:19:46 -0700
  • 6e559e9b5f Add MViT (Multi-Scale) V2 Ross Wightman 2022-08-17 15:12:31 -0700
  • 43aa84e861 Add 'fast' layer norm that doesn't cast to float32, support APEX LN impl for slight speed gain, update norm and act factories, tweak SE for ability to disable bias (needed by GCVit) Ross Wightman 2022-08-17 14:32:58 -0700
  • c486aa71f8 Add GCViT Ross Wightman 2022-08-17 14:29:18 -0700
  • fba6ecd39b Add EfficientFormer Ross Wightman 2022-08-17 14:08:53 -0700
  • ff4a38e2c3 Add PyramidVisionTransformerV2 Ross Wightman 2022-08-17 12:06:05 -0700
  • 5a40c6a3c4 Fix issue with torchvision's ImageNet Edoardo Debenedetti 2022-08-17 15:10:41 +0200
  • 1d8ada359a Add timm ConvNeXt 'atto' weights, change test resolution for FB ConvNeXt 224x224 weights, add support for different dw kernel_size v0.1-weights-morevit Ross Wightman 2022-08-15 17:56:08 -0700
  • 7c4682dc08 Update README.md Ross Wightman 2022-08-05 17:28:56 -0700
  • 2544d3b80f ConvNeXt pico, femto, and nano, pico, femto ols (overlapping stem) weights and model defs Ross Wightman 2022-08-05 17:05:50 -0700
  • 13565aad50 Add edgenext_base model def & weight link, update to improve ONNX export #1385 Ross Wightman 2022-08-05 16:58:34 -0700
  • 56596e4e84 jit trace comparisons snuck into torchscript part of validate.py, fixed Ross Wightman 2022-07-31 21:13:56 -0700
  • ed1057f1e7 Make EdgeNeXt onnx-exportable. #1385 kakukakujirori 2022-07-30 15:19:26 +0900
  • e91c2181b3 Fixes 1.0 crop reproducibility issue. #1383 alih 2022-07-29 11:09:13 -0700
  • 8ad4bdfa06 Allow ntuple to be used with string values Ross Wightman 2022-07-28 16:18:18 -0700
  • faae93e62d
         Fix typo in PositionalEncodingFourier #1381 Christoph Reich 2022-07-28 19:08:08 -0400
  • bd6d377c74 Merge remote-tracking branch 'origin/master' into bits_and_tpu Ross Wightman 2022-07-28 15:55:23 -0700
  • 7430a85d07 Update README, bump version to 0.6.8 Ross Wightman 2022-07-28 15:07:11 -0700
  • ec6a28830f Add DeiT-III 'medium' model defs and weights Ross Wightman 2022-07-28 15:03:20 -0700
  • 7cd4204a28 Add TPU TRC acknowledge v0.6.7 Ross Wightman 2022-07-27 14:07:37 -0700
  • 7d44d65bf5 Update README and changelogs Ross Wightman 2022-07-27 14:04:29 -0700
  • d875a1d3f6 version 0.6.7 Ross Wightman 2022-07-27 12:41:06 -0700
  • c865028c34 Update benchmark with latest model adds Ross Wightman 2022-07-27 12:40:22 -0700
  • 30bd1746c5 Improve csv table result processing for better sort when updating Ross Wightman 2022-07-27 12:26:52 -0700
  • e987e29036 Add convnext_nano and few cs3 models to existing results tables Ross Wightman 2022-07-27 12:26:34 -0700
  • 6f103a442b Add convnext_nano weights, 80.8 @ 224, 81.5 @ 288 Ross Wightman 2022-07-26 16:37:36 -0700
  • 4042a94f8f Add weights for two 'Edge' block (3x3->1x1) variants of CS3 networks. Ross Wightman 2022-07-26 16:36:04 -0700
  • c8f69e04a9
         Merge pull request #1365 from veritable-tech/fix-resize-pos-embed Ross Wightman 2022-07-24 21:03:01 -0700
  • 99af63ca92
         Merge pull request #1277 from lukasugar/patch-1 Ross Wightman 2022-07-24 15:35:48 -0700
  • 45c447fc15
         Merge pull request #1363 from Jasha10/patch-1 Ross Wightman 2022-07-24 15:34:14 -0700
  • 0b64117592 Take `no_emb_class` into account when calling `resize_pos_embed` #1365 Ceshine Lee 2022-07-24 19:11:45 +0800
  • 56c3a84db3
         Update type hint for `register_notrace_module` #1363 Jasha10 2022-07-22 16:59:55 -0500
  • d7b55a9429 Add gmacs and macts columns to inference benchmark (missed profile in initial run) Ross Wightman 2022-07-22 12:12:37 -0700
  • 1b278136c3 Change models with mean 0,0,0 std 1,1,1 from int to float for consistency as mentioned in #1355 Ross Wightman 2022-07-21 17:36:15 -0700
  • 909705e7ff Remove some redundant requires_grad=True from nn.Parameter in third party code Ross Wightman 2022-07-20 12:37:41 -0700
  • c5e0d1c700 Add dilation support to convnext, allows output_stride=8 and 16 use. Fix #1341 Ross Wightman 2022-07-19 17:52:10 -0700
  • 5e7d47ca10 Add pytorch 1.12 benchmark csv files w/ 0.6.6 code. Remove pytorch 1.10 results. Deciding whether to update 1.11 results or remove... Ross Wightman 2022-07-19 15:29:38 -0700
  • dc376e3676 Ensure all model entrypoint fn default to `pretrained=False` (a few didn't) Ross Wightman 2022-07-19 13:58:41 -0700
  • 23b102064a Add cs3sedarknet_x weights w/ 82.65 @ 288 top1. Add 2 cs3 edgenet models (w/ 3x3-1x1 block), remove aa from cspnet blocks (not needed) Ross Wightman 2022-07-19 13:21:20 -0700
  • 0dbd9352ce Add bulk_runner script and updates to benchmark.py and validate.py for better error handling in bulk runs (used for benchmark and validation result runs). Improved batch size decay stepping on retry... Ross Wightman 2022-07-18 18:01:39 -0700
  • 4547920f85
         Merge pull request #1354 from rwightman/fix_tests Ross Wightman 2022-07-17 21:37:16 -0700
  • 29afe79c8b Attempt to fix unit tests by removing subset of tests on mac runner #1354 fix_tests Ross Wightman 2022-07-17 14:55:47 -0700
  • 326ade2999 Add updated validation / test set results, benchmarks still running... Ross Wightman 2022-07-16 16:15:28 -0700
  • 92b91af3bb version 0.6.6 Ross Wightman 2022-07-15 17:00:18 -0700
  • 05313940e2 Add cs3darknet_x, cs3sedarknet_l, and darknetaa53 weights from TPU sessions. Move SE btwn conv1 & conv2 in DarkBlock. Improve SE/attn handling in Csp/DarkNet. Fix leaky_relu bug on older csp models. Ross Wightman 2022-07-15 16:53:51 -0700
  • 4283c0c478
         Merge pull request #1351 from nateraw/use-hf-hub-download Ross Wightman 2022-07-14 14:51:02 -0700
  • 51cca82aa1 👽 use hf_hub_download instead of cached_download #1351 nateraw 2022-07-14 16:41:45 -0400
  • 64942273ac
         Update helpers.py #1349 Andrey Gurevich 2022-07-14 12:01:39 +0300
  • 324a4e58b6 disable nvfuser for jit te/legacy modes (for PT 1.12+) Ross Wightman 2022-07-13 10:34:34 -0700
  • 6fe01993ad verions 0.8.x for bits_and_tpu branch Ross Wightman 2022-07-13 09:53:15 -0700
  • 1186fc9c73 Merge remote-tracking branch 'origin/master' into bits_and_tpu Ross Wightman 2022-07-12 17:14:35 -0700
  • f96da54eb1
         Merge pull request #1345 from mmaaz60/patch-1 Ross Wightman 2022-07-11 20:34:46 -0700
  • 4e7ffe5043
         Update README.md #1345 Muhammad Maaz 2022-07-12 05:08:34 +0400
  • 2898cf6e41 version 0.6.5 for pypi release v0.6.5 Ross Wightman 2022-07-10 16:43:23 -0700
  • 8f2df42aa5 hack ln implementation #1340 Natalia Gimelshein 2022-07-09 23:20:31 +0000
  • 66393d472f Update README.md Ross Wightman 2022-07-08 12:21:23 -0700
  • a45b4bce9a x and xx small edgenext models do benefit from larger test input size Ross Wightman 2022-07-08 10:53:27 -0700
  • a8e34051c1 Unbreak gamma remap impacting beit checkpoint load, version bump to 0.6.4 Ross Wightman 2022-07-07 23:07:43 -0700
  • 1ccce50d48
         Merge pull request #1327 from rwightman/edgenext_csp_and_more Ross Wightman 2022-07-07 22:06:47 -0700
  • 1c5cb819f9 bump version to 0.6.3 before merge #1327 edgenext_csp_and_more Ross Wightman 2022-07-07 22:05:56 -0700
  • a1cb25066e Add edgnext_small_rw weights trained with swin like recipe. Better than original 'small' but not the recent 'USI' distilled weights. Ross Wightman 2022-07-07 22:02:57 -0700
  • 7c7ecd2492 Add --use-train-size flag to force use of train input_size (over test input size) for validation. Default test-time pooling to use train input size (fixes issues). Ross Wightman 2022-07-07 22:01:24 -0700
  • ce65a7b29f Update vit_relpos w/ some additional weights, some cleanup to match recent vit updates, more MLP log coord experiments. Ross Wightman 2022-07-07 21:33:25 -0700
  • 58621723bd Add CrossStage3 DarkNet (cs3) weights Ross Wightman 2022-07-07 17:43:38 -0700
  • 9be0c84715 Change set -> dict w/ None keys for dataset split synonym search, so always consistent if more than 1 exists. Fix #1224 Ross Wightman 2022-07-07 15:33:53 -0700
  • 4670d375c6 Reorg benchmark.py import Ross Wightman 2022-07-07 15:21:29 -0700
  • 2456223052
         Merge pull request #1336 from xwang233/add-local-rank Ross Wightman 2022-07-07 15:20:30 -0700
  • 500c190860 Add --aot-autograd (functorch efficient mem fusion) support to validate.py Ross Wightman 2022-07-07 15:15:25 -0700
  • 28e0152043 Add --no-retry flag to benchmark.py to skip batch_size decay and retry on error. Fix #1226. Update deepspeed profile usage for latest DS releases. Fix # 1333 Ross Wightman 2022-07-07 15:13:06 -0700
  • 11060f84c5 make train.py compatible with torchrun #1336 Xiao Wang 2022-07-07 14:44:55 -0700
  • db0cee9910 Refactor cspnet configuration using dataclasses, update feature extraction for new cs3 variants. Ross Wightman 2022-07-07 14:43:27 -0700
  • eca09b8642 Add MobileVitV2 support. Fix #1332. Move GroupNorm1 to common layers (used in poolformer + mobilevitv2). Keep ol custom ConvNeXt LayerNorm2d impl as LayerNormExp2d for reference. Ross Wightman 2022-07-07 14:41:01 -0700
  • 06307b8b41 Remove experimental downsample in block support in ConvNeXt. Experiment further before keeping it in. Ross Wightman 2022-07-07 14:37:58 -0700
  • bfc0dccb0e Improve image extension handling, add methods to modify / get defaults. Fix #1335 fix #1274. Ross Wightman 2022-07-07 14:23:20 -0700
  • 7d4b3807d5 Support DeiT-3 (Revenge of the ViT) checkpoints. Add non-overlapping (w/ class token) pos-embed support to vit. Ross Wightman 2022-07-04 22:25:22 -0700
  • eb143c3e35
         export.py added to export timm models to onnx swenkel 2022-07-03 21:44:47 +0300
  • d0c5bd5722 Rename cs2->cs3 for darknets. Fix features_only for cs3 darknets. Ross Wightman 2022-07-03 08:32:41 -0700
  • d765305821 Remove first_conv for resnetaa50 def Ross Wightman 2022-07-02 15:56:17 -0700
  • dd9b8f57c4 Add feature_info to edgenext for features_only support, hopefully fix some fx / test errors Ross Wightman 2022-07-02 15:20:45 -0700
  • 377e9bfa21 Add TPU trained darknet53 weights. Add mising pretrain_cfg for some csp/darknet models. Ross Wightman 2022-07-02 15:18:52 -0700
  • c170ba3173 Add weights for resnet10t, resnet14t, and resnetaa50 models. Fix #1314 Ross Wightman 2022-07-02 15:18:06 -0700
  • 188c194b0f Left some experiment stem code in convnext by mistake Ross Wightman 2022-07-02 15:17:28 -0700
  • 70d6d2c484 support test_crop_size in data config resolve Ross Wightman 2022-07-02 15:17:05 -0700
  • 6064d16a2d Add initial EdgeNeXt import. Significant cleanup / reorg (like ConvNeXt). Fix #1320 * edgenext refactored for torchscript compat, stage base organization * slight refactor of ConvNeXt to match some EdgeNeXt additions * remove use of funky LayerNorm layer in ConvNeXt and just use nn.LayerNorm and LayerNorm2d (permute) Ross Wightman 2022-07-01 15:16:41 -0700
  • 7a9c6811c9 Add eps arg to LayerNorm2d, add 'tf' (tensorflow) variant of trunc_normal_ that applies scale/shift after sampling (instead of needing to move a/b) Ross Wightman 2022-07-01 15:15:39 -0700
  • 82c311d082 Add more experimental darknet and 'cs2' darknet variants (different cross stage setup, closer to newer YOLO backbones) for train trials. Ross Wightman 2022-07-01 15:14:01 -0700
  • a050fde5cd Add resnet10t (basic block) and resnet14t (bottleneck) with 1,1,1,1 repeats Ross Wightman 2022-07-01 15:03:28 -0700
  • 34f382f8f6 move dataconfig before script, scripting killing metadata now (PyTorch 1.12? just nvfuser?) Ross Wightman 2022-07-01 14:50:36 -0700
  • e9ff104ac9
         (chore) Remove unused import `os` in benchmark.py #1322 phi9t 2022-06-28 13:27:44 -0700
  • beef62e7ab
         Merge pull request #1317 from rwightman/fixes-syncbn_pretrain_cfg_resolve Ross Wightman 2022-06-25 07:37:59 -0700
  • e6d7df40ec no longer a point using kwargs for pretrain_cfg resolve, just pass explicit arg #1317 fixes-syncbn_pretrain_cfg_resolve Ross Wightman 2022-06-24 21:32:44 -0700
  • a29fba307d disable dist_bn when sync_bn active Ross Wightman 2022-06-24 21:30:17 -0700
  • 07d0c4ae96 Improve repr for DropPath module Ross Wightman 2022-06-24 14:58:15 -0700
  • e27c16b8a0 Remove unecessary code for synbn guard Ross Wightman 2022-06-24 14:57:42 -0700
  • 0da3c9ebbf Remove SiLU layer in default args that breaks import on old old PyTorch Ross Wightman 2022-06-24 14:56:58 -0700