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

Commit Graph

  • ba2ca4b464 One codepath for stdconv, switch layernorm to batchnorm so gain included. Tweak epsilon values for nfnet, resnetv2, vit hybrid. Ross Wightman 2021-06-12 12:27:43 -0700
  • 07fb05cc3d Update results csv files Ross Wightman 2021-06-09 22:33:05 -0700
  • b79dfd4fc2
         Merge pull request #693 from SamuelGabriel/patch-1 Ross Wightman 2021-06-09 14:30:05 -0700
  • 7c19c35d9f
         Global instead of local rank. #693 SamuelGabriel 2021-06-09 19:11:58 +0200
  • b7a568f065 Fix torchscript issue in bat Ross Wightman 2021-06-08 23:19:51 -0700
  • d17b374f0f Minimum input_size needed to be higher Ross Wightman 2021-06-08 21:31:39 -0700
  • b3b90d944d Add min_input_size to bat_resnext to prevent test breakage. Ross Wightman 2021-06-08 17:32:08 -0700
  • 758c4438a7 Update README.md Ross Wightman 2021-06-08 15:19:11 -0700
  • d413eef1bf Add ResMLP-24 model weights that I trained in PyTorch XLA on TPU-VM. 79.2 top-1. Ross Wightman 2021-06-08 14:22:05 -0700
  • 10d8fa4620 Add gc and bat attention resnext26ts variants to byob for test. Ross Wightman 2021-06-08 14:21:07 -0700
  • 2f5ed2dec1 Update `init_values` const for 24 and 36 layer ResMLP models Ross Wightman 2021-06-07 17:15:04 -0700
  • 8e4ac3549f All ScaledStdConv and StdConv uses default to using F.layernorm so that they work with PyTorch XLA. eps value tweaking is a WIP. Ross Wightman 2021-06-07 17:14:19 -0700
  • 2a63d0246b Post merge cleanup Ross Wightman 2021-06-07 14:38:30 -0700
  • 45dec179e5
         Merge pull request #681 from lmk123568/master Ross Wightman 2021-06-07 14:10:53 -0700
  • 4907f8f70d
         Merge pull request #685 from dyhan0920/master Ross Wightman 2021-06-07 14:08:45 -0700
  • ded1671483 Fix stochastic depth working only with a shortcut #685 Dongyoon Han 2021-06-07 23:08:55 +0900
  • 847b4af144
         Update README.md Ross Wightman 2021-06-06 12:35:54 -0700
  • b87d98b238
         Update convit.py #681 Mike 2021-06-06 17:58:31 +0800
  • 0b718a82c7
         one can implement CONTAINER-PAM and obtain a +0.5 improvement on ImageNet top-1 accuracy ##from :https://arxiv.org/pdf/2106.01401.pdf #679 小菜 2021-06-05 16:58:36 +0800
  • 5c5cadfe4c
         Update README.md Ross Wightman 2021-06-04 14:44:07 -0700
  • ee2b8f49ee
         Update README.md Ross Wightman 2021-06-04 14:41:29 -0700
  • cc870df7b8
         Update README.md Ross Wightman 2021-06-04 14:23:34 -0700
  • 6b2d9c2660 Another bits/README.md update Ross Wightman 2021-06-04 13:08:29 -0700
  • c3db5f5801 Worker hack for TFDS eval, add TPU env var setting. Ross Wightman 2021-06-04 13:00:52 -0700
  • f411724de4 Fix checkpoint delete issue. Add README about bits and initial Pytorch XLA usage on TPU-VM. Add some FIXMEs and fold train_cfg into train_state by default. Ross Wightman 2021-06-04 12:49:53 -0700
  • b57a03bd0d Merge branch 'master' into bits_and_tpu Ross Wightman 2021-06-03 17:51:20 -0700
  • 91ab0b6ce5 Add proper TrainState checkpoint save/load. Some reorg/refactoring and other cleanup. More to go... Ross Wightman 2021-06-03 17:49:40 -0700
  • 54a6cca27a
         Merge pull request #668 from rwightman/more_attn Ross Wightman 2021-05-31 15:52:24 -0700
  • 02320c3e3d Bump version to 0.4.11 #668 more_attn Ross Wightman 2021-05-31 15:41:51 -0700
  • bda8ab015a Remove min channels for SelectiveKernel, divisor should cover cases well enough. Ross Wightman 2021-05-31 15:38:56 -0700
  • a27f4aec4a Missed args for skresnext w/ refactoring. Ross Wightman 2021-05-31 14:06:34 -0700
  • 307a935b79 Add non-local and BAT attention. Merge attn and self-attn factories into one. Add attention references to README. Add mlp 'mode' to ECA. Ross Wightman 2021-05-31 13:18:11 -0700
  • aebc5b58c9 fix some bugs #659 lixinjie 2021-05-31 15:18:46 +0800
  • 17dc47c8e6 Missed comma in test filters. Ross Wightman 2021-05-30 22:00:43 -0700
  • 56bd5822f1 fix lixinjie 2021-05-31 12:51:25 +0800
  • 34522097b1 See if we can use tcmalloc in test runner Ross Wightman 2021-05-30 21:12:10 -0700
  • 8bf63b6c6c Able to use other attn layer in EfficientNet now. Create test ECA + GC B0 configs. Make ECA more configurable. Ross Wightman 2021-05-30 12:47:02 -0700
  • 8d9ebe3788 fix some bugs lixinjie 2021-05-31 00:36:13 +0800
  • bd996241f6 fix some bugs lixinjie 2021-05-30 21:25:28 +0800
  • bcec14d3b5 Bring EfficientNet SE layer in line with others, pull se_ratio outside of blocks. Allows swapping w/ other attn layers. Ross Wightman 2021-05-29 23:41:38 -0700
  • 041a85fe7b fix some bugs to support segmentation task lixinjie 2021-05-30 12:55:21 +0800
  • f8a16f5267
         Merge branch 'master' into master abcdvzz 2021-05-29 21:26:53 +0800
  • 9611458e19 Throw in some FBNetV3 code I had lying around, some refactoring of SE reduction channel calcs for all EffNet archs. Ross Wightman 2021-05-28 20:47:24 -0700
  • 01b9108619 Merge branch 'master' into more_attn Ross Wightman 2021-05-28 11:09:37 -0700
  • d7bab8a6c5 Fix strict flag change for checkpoint load. Ross Wightman 2021-05-28 09:54:50 -0700
  • 02f9d4bc34 Add weights for resnet51q model, add 61q def. Ross Wightman 2021-05-28 09:53:16 -0700
  • c0c03c4e97 fix args name bug #669 John Krafcik 2021-05-28 11:39:44 +0800
  • f615474be3 Fix broken test, repvgg block doesn't have attn_last attr. Ross Wightman 2021-05-27 18:12:22 -0700
  • 742c2d5247 Add Gather-Excite and Global Context attn modules. Refactor existing SE-like attn for consistency and refactor byob/byoanet for less redundancy. Ross Wightman 2021-05-27 18:03:29 -0700
  • 9c78de8c02 Fix #661, move hardswish out of default args for LeViT. Enable native torch support for hardswish, hardsigmoid, mish if present. Ross Wightman 2021-05-26 15:28:42 -0700
  • 07d952c7a7
         Merge pull request #637 from rwightman/levit_visformer_rednet Ross Wightman 2021-05-25 14:27:06 -0700
  • 7f368782b7
         Merge pull request #660 from petervandenabeele/readme_fix_typos Ross Wightman 2021-05-25 14:26:51 -0700
  • 5db7452173 Fix visformer in_chans stem handling #637 Ross Wightman 2021-05-25 14:11:36 -0700
  • 51c432150a README: fix simple typos #660 Peter Vandenabeele 2021-05-25 22:42:44 +0200
  • fd92ba0de8 Filter large vit models from torchscript tests Ross Wightman 2021-05-25 12:52:07 -0700
  • 318360c3f9 Update README.md before merge. Bump version to 0.4.10 Ross Wightman 2021-05-25 12:25:53 -0700
  • 99d97e0d67 Hopefully the last test update for this PR... Ross Wightman 2021-05-25 11:10:17 -0700
  • 11ae795e99 Redo LeViT attention bias caching in a way that works with both torchscript and DataParallel Ross Wightman 2021-05-25 10:15:32 -0700
  • d400f1dbdd Filter test models before creation for backward/torchscript tests Ross Wightman 2021-05-25 10:14:45 -0700
  • 43180b1341 update twins.py to support segmentation task 李鑫杰 2021-05-25 18:07:50 +0800
  • c4572cc5aa Add Visformer-small weighs, tweak torchscript jit test img size. Ross Wightman 2021-05-24 22:50:12 -0700
  • 83487e2a0d Lower max backward size for tests. Ross Wightman 2021-05-24 21:36:56 -0700
  • bfc72f75d3 Expand scope of testing for non-std vision transformer / mlp models. Some related cleanup and create fn cleanup for all vision transformer and mlp models. More CoaT weights. Ross Wightman 2021-05-24 21:13:26 -0700
  • 18bf520ad1 Add eca_nfnet_l2/l3 defs for future training Ross Wightman 2021-05-22 21:55:37 -0700
  • f45de37690 Merge branch 'master' into levit_visformer_rednet Ross Wightman 2021-05-22 16:34:31 -0700
  • 23c18a33e4 Add efficientnetv2_rw_m weights trained in PyTorch. 84.8 top-1 @ 416 test. 53M params. Ross Wightman 2021-05-21 21:16:25 -0700
  • 5b9c69e80a Add basic training resume based on legacy code Ross Wightman 2021-05-21 18:08:06 -0700
  • c2ba229d99 Prep for effcientnetv2_rw_m model weights that started training before official release.. Ross Wightman 2021-05-21 17:47:49 -0700
  • 22f7c6760f Merge branch 'alexander-soare-master' Ross Wightman 2021-05-21 17:21:40 -0700
  • 30b9880d06 Minor adjustment, mutable default arg, extra check of valid len... Ross Wightman 2021-05-21 17:20:33 -0700
  • be0abfbcce Merge branch 'master' of https://github.com/alexander-soare/pytorch-image-models into alexander-soare-master Ross Wightman 2021-05-21 17:10:11 -0700
  • d4c00d6150 Merge branch 'amaarora-convit' Ross Wightman 2021-05-21 17:04:56 -0700
  • b7de82e835 ConViT cleanup, fix torchscript, bit of reformatting, reuse existing layers. Ross Wightman 2021-05-21 17:04:23 -0700
  • 306c86b668 Merge branch 'convit' of https://github.com/amaarora/pytorch-image-models into amaarora-convit Ross Wightman 2021-05-21 16:27:10 -0700
  • 6e04da0c54 Merge branch 'abcdvzz-master' Ross Wightman 2021-05-21 16:24:46 -0700
  • a569635045 Update twin weights to a copy in GitHub releases for faster dl. Tweak model class comment. Ross Wightman 2021-05-21 16:23:14 -0700
  • be99eef9c1 Remove redundant code, cleanup, fix torchscript. Ross Wightman 2021-05-20 23:38:35 -0700
  • 5ab372a3ec Merge branch 'master' of https://github.com/abcdvzz/pytorch-image-models into abcdvzz-master Ross Wightman 2021-05-20 23:37:50 -0700
  • 50d6aab0ef Add convit to non-std filters as vit_ #653 Aman Arora 2021-05-21 03:46:47 +0000
  • 5db1eb6ba5 Add defaults Aman Arora 2021-05-21 02:11:20 +0000
  • 9b97753f9c Merge branch 'convit' of https://github.com/amaarora/pytorch-image-models into convit Aman Arora 2021-05-21 01:12:10 +0000
  • 1633317489 update tests and exclude convit_base Aman Arora 2021-05-21 01:11:56 +0000
  • 8b1f2e8e1f remote unused matplotlib import Aman Arora 2021-05-20 23:42:42 +0000
  • 40c506ba1e Add ConViT Aman Arora 2021-05-20 23:17:28 +0000
  • b4ebf9263e
         Merge pull request #651 from mrT23/master v0.1-vt3p-weights Ross Wightman 2021-05-20 07:55:38 -0700
  • 7976019864 extend positional embedding resizing functionality to tnt #652 Alexander Soare 2021-05-20 11:55:48 +0100
  • 8086943b6f allow resize positional embeddings to non-square grid Alexander Soare 2021-05-20 11:27:58 +0100
  • dc1a4efd28 mixer_b16_224_miil, mixer_b16_224_miil_in21k models #651 talrid 2021-05-20 10:35:50 +0300
  • 709d7c07e8 Merge branch 'rwightman:master' into master mrT23 2021-05-20 10:23:20 +0300
  • 240e667746 Revert "mixer_b16_224_miil" talrid 2021-05-20 10:23:07 +0300
  • 4210d922d2 Merge branch 'master' into bits_and_tpu Ross Wightman 2021-05-19 23:11:47 -0700
  • d046498e0b update test_models.py #645 李鑫杰 2021-05-20 11:20:39 +0800
  • 7b799c4e79 add latest code 李鑫杰 2021-05-20 11:15:49 +0800
  • cf4ce2f15b
         Merge pull request #647 from rwightman/more_mlp Ross Wightman 2021-05-19 12:47:12 -0700
  • 72ca831dd4 Back to using strings for the enum translation, forgot about import dep Ross Wightman 2021-05-19 12:01:53 -0700
  • b3b503c13d
         Merge pull request #626 from cvcomar/master Ross Wightman 2021-05-19 11:51:59 -0700
  • 6d81374b88 Update tests for new mlp models #647 Ross Wightman 2021-05-19 11:09:42 -0700
  • 5bcf686cb0 mixer_b16_224_miil talrid 2021-05-19 20:51:10 +0300
  • d5af752117 Add preliminary gMLP and ResMLP impl to Mlp-Mixer Ross Wightman 2021-05-19 09:55:05 -0700
  • cbd4ee737f Fix model init for XLA, remove some prints. Ross Wightman 2021-05-18 16:54:36 -0700