e91c2181b3
Fixes 1.0 crop reproducibility issue.
...
transforms.CenterCrop automatically converts image size into a tuple, but transforms.Resize does not.
When running models on ImageNet at 384x384 with 1.0 crop_pct, this will result in transforms.Resize(384).
Some methods (i.e. ConvNeXt, Swin) implement this with transforms.Resize((384, 384)).
As a result their reported scores on ImageNet at 384 are not reproducible with timm's validation script.
3 years ago
7430a85d07
Update README, bump version to 0.6.8
3 years ago
ec6a28830f
Add DeiT-III 'medium' model defs and weights
3 years ago
7cd4204a28
Add TPU TRC acknowledge
3 years ago
7d44d65bf5
Update README and changelogs
3 years ago
d875a1d3f6
version 0.6.7
3 years ago
c865028c34
Update benchmark with latest model adds
3 years ago
30bd1746c5
Improve csv table result processing for better sort when updating
3 years ago
e987e29036
Add convnext_nano and few cs3 models to existing results tables
3 years ago
6f103a442b
Add convnext_nano weights, 80.8 @ 224, 81.5 @ 288
3 years ago
4042a94f8f
Add weights for two 'Edge' block (3x3->1x1) variants of CS3 networks.
3 years ago
c8f69e04a9
Merge pull request #1365 from veritable-tech/fix-resize-pos-embed
...
Take `no_emb_class` into account when calling `resize_pos_embed`
3 years ago
99af63ca92
Merge pull request #1277 from lukasugar/patch-1
...
Add missing output in Feature extraction docs
3 years ago
45c447fc15
Merge pull request #1363 from Jasha10/patch-1
...
Update type hint for `register_notrace_module`
3 years ago
0b64117592
Take `no_emb_class` into account when calling `resize_pos_embed`
3 years ago
56c3a84db3
Update type hint for `register_notrace_module`
...
register_notrace_module is used to decorate types (i.e. subclasses of nn.Module).
It is not called on module instances.
3 years ago
d7b55a9429
Add gmacs and macts columns to inference benchmark (missed profile in initial run)
3 years ago
1b278136c3
Change models with mean 0,0,0 std 1,1,1 from int to float for consistency as mentioned in #1355
3 years ago
909705e7ff
Remove some redundant requires_grad=True from nn.Parameter in third party code
3 years ago
c5e0d1c700
Add dilation support to convnext, allows output_stride=8 and 16 use. Fix #1341
3 years ago
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...
3 years ago
dc376e3676
Ensure all model entrypoint fn default to `pretrained=False` (a few didn't)
3 years ago
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)
3 years ago
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...
3 years ago
4547920f85
Merge pull request #1354 from rwightman/fix_tests
...
Attempting to fix unit test failures...
3 years ago
29afe79c8b
Attempt to fix unit tests by removing subset of tests on mac runner
3 years ago
326ade2999
Add updated validation / test set results, benchmarks still running...
3 years ago
92b91af3bb
version 0.6.6
3 years ago
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.
3 years ago
4283c0c478
Merge pull request #1351 from nateraw/use-hf-hub-download
...
Use hf_hub_download instead of cached_download
3 years ago
51cca82aa1
👽 use hf_hub_download instead of cached_download
3 years ago
324a4e58b6
disable nvfuser for jit te/legacy modes (for PT 1.12+)
3 years ago
f96da54eb1
Merge pull request #1345 from mmaaz60/patch-1
...
Update README.md
3 years ago
4e7ffe5043
Update README.md
3 years ago
2898cf6e41
version 0.6.5 for pypi release
3 years ago
66393d472f
Update README.md
3 years ago
a45b4bce9a
x and xx small edgenext models do benefit from larger test input size
3 years ago
a8e34051c1
Unbreak gamma remap impacting beit checkpoint load, version bump to 0.6.4
3 years ago
1ccce50d48
Merge pull request #1327 from rwightman/edgenext_csp_and_more
...
EdgeNeXt, additional DarkNets, and more
3 years ago
1c5cb819f9
bump version to 0.6.3 before merge
3 years ago
a1cb25066e
Add edgnext_small_rw weights trained with swin like recipe. Better than original 'small' but not the recent 'USI' distilled weights.
3 years ago
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).
3 years ago
ce65a7b29f
Update vit_relpos w/ some additional weights, some cleanup to match recent vit updates, more MLP log coord experiments.
3 years ago
58621723bd
Add CrossStage3 DarkNet (cs3) weights
3 years ago
9be0c84715
Change set -> dict w/ None keys for dataset split synonym search, so always consistent if more than 1 exists. Fix #1224
3 years ago
4670d375c6
Reorg benchmark.py import
3 years ago
2456223052
Merge pull request #1336 from xwang233/add-local-rank
...
Make train.py compatible with torchrun
3 years ago
500c190860
Add --aot-autograd (functorch efficient mem fusion) support to validate.py
3 years ago
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
3 years ago
11060f84c5
make train.py compatible with torchrun
3 years ago
alih