Commits

Branch selector

 vision_transformer

User selector

All users

Datepicker

All time

Commit History

Commits on Oct 13, 2020

• Update README, ensure vit excluded from all tests (not ready)

  rwightman

  committed
  be53107

  Copy full SHA for be53107

• Initial Vision Transformer impl w/ patch and hybrid variants. Refactor tuple helpers.

  rwightman

  committed
  f31933c

  Copy full SHA for f31933c

• Default to old checkpoint format for now, still want compatibility with older torch ver for released models

  rwightman

  committed
  9305313

  Copy full SHA for 9305313

• Add model based wd skip support. Improve cross version compat of optimizer factory. Fix #247

  rwightman

  committed
  a4d8fea

  Copy full SHA for a4d8fea

Commits on Oct 10, 2020

• Add Adafactor and Adahessian optimizers, cleanup optimizer arg passing, add gradient clipping support.

  rwightman

  committed
  80078c4

  Copy full SHA for 80078c4

Commits on Oct 2, 2020

• Add missing leaky_relu layer factory defn, update Apex/Native loss scaler interfaces to support unscaled grad clipping. Bump ver to 0.2.2 for pending release.

  rwightman

  committed
  fcb6258

  Copy full SHA for fcb6258

• Merge pull request #244 from hollance/master

  Show description for 186075e
  rwightman

  authored
  186075e

  Copy full SHA for 186075e

• don't forget this file

  hollance

  committed
  f04bdc8

  Copy full SHA for f04bdc8

• test_time_pool would be set to a non-False value even if test-time pooling is not available

  hollance

  committed
  8ffdc59

  Copy full SHA for 8ffdc59

Commits on Sep 25, 2020

• Missed moving some seresnet -> legacy in sotabench. Check sotabench cache.

  rwightman

  committed
  4be5b51

  Copy full SHA for 4be5b51

• Add DropPath (stochastic depth) to ReXNet and VoVNet. RegNet DropPath impl tweak and dedupe se args.

  rwightman

  committed
  e8e2d9c

  Copy full SHA for e8e2d9c

Commits on Sep 24, 2020

• More models in sotabench, more control over sotabench run, dataset filename extraction consistency

  rwightman

  committed
  e8ca458

  Copy full SHA for e8ca458

• Add EfficientNet-EdgeTPU-M (efficientnet_em) model trained natively in PyTorch. More sotabench fiddling.

  rwightman

  committed
  9c40653

  Copy full SHA for 9c40653

Commits on Sep 19, 2020

• Another sotabench.py debug iter

  rwightman

  committed
  3681c5c

  Copy full SHA for 3681c5c

• Sotabench debugging

  rwightman

  committed
  0802985

  Copy full SHA for 0802985

Commits on Sep 18, 2020

• Add ResNet weights. 80.5 (top-1) ResNet-50-D, 77.1 ResNet-34-D, 72.7 ResNet-18-D.

  rwightman

  committed
  c40384f

  Copy full SHA for c40384f

Commits on Sep 11, 2020

• Merge pull request #237 from rwightman/utils_cleanup

  Show description for e39bf6e
  rwightman

  authored
  e39bf6e

  Copy full SHA for e39bf6e

Commits on Sep 8, 2020

• More flexible mixup mode, add 'half' mode.

  rwightman

  committed
  47a7b3b

  Copy full SHA for 47a7b3b

Commits on Sep 7, 2020

• Reorg of utils into separate modules

  rwightman

  committed
  532e3b4

  Copy full SHA for 532e3b4

Commits on Sep 3, 2020

• Update README.md

  rwightman

  authored
  9ce42d5

  Copy full SHA for 9ce42d5

• Update README.md

  rwightman

  authored
  0729dbe

  Copy full SHA for 0729dbe

• Updated README, add wide_resnet50_2 and seresnext50_32x4d weights

  rwightman

  committed
  33f8a1b

  Copy full SHA for 33f8a1b

• Merge pull request #233 from rwightman/torchamp

  Show description for 5247eb3
  rwightman

  authored
  5247eb3

  Copy full SHA for 5247eb3

Commits on Sep 2, 2020

• Add global_pool (--gp) arg changes to allow passing 'fast' easily for train/validate to avoid channels_last issue with AdaptiveAvgPool

  rwightman

  committed
  751b0bb

  Copy full SHA for 751b0bb

• Cleanup Apex vs native AMP scaler state save/load. Cleanup CheckpointSaver a bit.

  rwightman

  committed
  9c297ec

  Copy full SHA for 9c297ec

• Add 'fast' global pool option, remove redundant SEModule from tresnet, normal one is now 'fast'

  rwightman

  committed
  80c9d9c

  Copy full SHA for 80c9d9c

• hrnet features_only pretrained weight loading issue. Fix #232.

  rwightman

  committed
  90a01f4

  Copy full SHA for 90a01f4

Commits on Sep 1, 2020

• AdaptiveAvgPool2d -> mean((2,3)) for all SE/attn layers to avoid NaN with AMP + channels_last layout. See https://github.com/pytorch/pytorch/issues/43992

  rwightman

  committed
  110a7c4

  Copy full SHA for 110a7c4

• Improve torch amp support and add channels_last support for train/validate scripts

  rwightman

  committed
  c2cd1a3

  Copy full SHA for c2cd1a3

Commits on Aug 31, 2020

• Merge branch 'master' of https://github.com/tgisaturday/pytorch-image-models into torchamp

  rwightman

  committed
  1d34a0a

  Copy full SHA for 1d34a0a

Commits on Aug 26, 2020

• Update README.md

  rwightman

  authored
  6d158ad

  Copy full SHA for 6d158ad

Commits on Aug 22, 2020

• fix save_checkpoint bug with native amp

  tgisaturday

  committed
  5f563ca

  Copy full SHA for 5f563ca

• add support for native torch AMP in torch 1.6

  tgisaturday

  committed
  d98967e

  Copy full SHA for d98967e

Commits on Aug 18, 2020

• Fix MobileNetV3 crash with global_pool='', output consistent with other models but not equivalent due to efficient head.

  rwightman

  committed
  470220b

  Copy full SHA for 470220b

Commits on Aug 14, 2020

• Fix a silly bug in Sample version of EvoNorm missing x* part of swish, update EvoNormBatch to accumulated unbiased variance.

  rwightman

  committed
  fc8b8af

  Copy full SHA for fc8b8af

Pagination

Previous

Next