Code

Refactoring of ImageProcessorFast#35069

Code

Merged

yonigozlan merged 59 commits intohuggingface:mainhuggingface/transformers:mainfrom yonigozlan:improve-fast-image-processor-baseyonigozlan/transformers:improve-fast-image-processor-baseCopy head branch name to clipboardFeb 4, 2025

+4,072-2,269

Commits

Commits on Jan 16, 2025

• add init and base image processing functions

  yonigozlan

  committed
  2f00f0c

  Copy full SHA for 2f00f0c

• add add_fast_image_processor to transformers-cli

  yonigozlan

  committed
  cfadb72

  Copy full SHA for cfadb72

• add working fast image processor clip

  yonigozlan

  committed
  2cd73cb

  Copy full SHA for 2cd73cb

• add fast image processor to doc, working tests

  yonigozlan

  committed
  932bd68

  Copy full SHA for 932bd68

• remove "to be implemented" SigLip

  yonigozlan

  committed
  23d79ce

  Copy full SHA for 23d79ce

• fix unprotected import

  yonigozlan

  committed
  3f2d8a6

  Copy full SHA for 3f2d8a6

• fix unprotected vision import

  yonigozlan

  committed
  6a9d332

  Copy full SHA for 6a9d332

• update ViTImageProcessorFast

  yonigozlan

  committed
  a1e2663

  Copy full SHA for a1e2663

• increase threshold slow fast ewuivalence

  yonigozlan

  committed
  fa74e7e

  Copy full SHA for fa74e7e

• add fast img blip

  yonigozlan

  committed
  9dbd765

  Copy full SHA for 9dbd765

• add fast class in tests with cli

  yonigozlan

  committed
  d39ff52

  Copy full SHA for d39ff52

• improve cli

  yonigozlan

  committed
  f609730

  Copy full SHA for f609730

• add fast image processor convnext

  yonigozlan

  committed
  8f7774d

  Copy full SHA for 8f7774d

• add LlavaPatchingMixin and fast image processor for llava_next and llava_onevision

  yonigozlan

  committed
  809e1f0

  Copy full SHA for 809e1f0

• add device kwarg to ImagesKwargs for fast processing on cuda

  yonigozlan

  committed
  f6e6cc2

  Copy full SHA for f6e6cc2

• cleanup

  yonigozlan

  committed
  e1ce148

  Copy full SHA for e1ce148

• fix unprotected import

  yonigozlan

  committed
  a24d89c

  Copy full SHA for a24d89c

• group images by sizes and add batch processing

  yonigozlan

  committed
  522e200

  Copy full SHA for 522e200

• Add batch equivalence tests, skip when center_crop is used

  yonigozlan

  committed
  deefc5a

  Copy full SHA for deefc5a

• cleanup

  yonigozlan

  committed
  6a2478e

  Copy full SHA for 6a2478e

• update init and cli

  yonigozlan

  committed
  7d76305

  Copy full SHA for 7d76305

• fix-copies

  yonigozlan

  committed
  142ed25

  Copy full SHA for 142ed25

• refactor convnext, cleanup base

  yonigozlan

  committed
  75bf56f

  Copy full SHA for 75bf56f

• fix

  yonigozlan

  committed
  de1fa18

  Copy full SHA for de1fa18

• remove patching mixins, add piped torchvision transforms for ViT

  yonigozlan

  committed
  2ffc41d

  Copy full SHA for 2ffc41d

• fix unbatched processing

  yonigozlan

  committed
  b524406

  Copy full SHA for b524406

• fix f strings

  yonigozlan

  committed
  9c2e2a4

  Copy full SHA for 9c2e2a4

• protect imports

  yonigozlan

  committed
  8c773e0

  Copy full SHA for 8c773e0

• change llava onevision to class transforms (test)

  yonigozlan

  committed
  90fceba

  Copy full SHA for 90fceba

• fix convnext

  yonigozlan

  committed
  e878bdd

  Copy full SHA for e878bdd

• improve formatting (following Pavel review)

  yonigozlan

  committed
  57acb7e

  Copy full SHA for 57acb7e

• fix handling device arg

  yonigozlan

  committed
  2a25104

  Copy full SHA for 2a25104

• improve cli

  yonigozlan

  committed
  4784fc8

  Copy full SHA for 4784fc8

• fix

  yonigozlan

  committed
  3ccd291

  Copy full SHA for 3ccd291

• fix inits

  yonigozlan

  committed
  053cdcb

  Copy full SHA for 053cdcb

Commits on Jan 21, 2025

• Merge remote-tracking branch 'upstream/main' into improve-fast-image-processor-base

  yonigozlan

  committed
  1b45e6e

  Copy full SHA for 1b45e6e

• Add distinction between preprocess and _preprocess, and support for arbitrary kwargs through valid_extra_kwargs

  yonigozlan

  committed
  9246945

  Copy full SHA for 9246945

Commits on Jan 22, 2025

• uniformize qwen2_vl fast

  yonigozlan

  committed
  6ccd230

  Copy full SHA for 6ccd230

• fix docstrings

  yonigozlan

  committed
  c4b8389

  Copy full SHA for c4b8389

• add add fast image processor llava

  yonigozlan

  committed
  e5c1e01

  Copy full SHA for e5c1e01

• remove min_pixels max_pixels from accepted size

  yonigozlan

  committed
  aef2fb4

  Copy full SHA for aef2fb4

• nit

  yonigozlan

  committed
  7078a14

  Copy full SHA for 7078a14

• nit

  yonigozlan

  committed
  aa94873

  Copy full SHA for aa94873

Commits on Jan 28, 2025

• refactor fast image processors docstrings

  yonigozlan

  committed
  13a125b

  Copy full SHA for 13a125b

• Merge remote-tracking branch 'upstream/main' into improve-fast-image-processor-base

  yonigozlan

  committed
  8adb893

  Copy full SHA for 8adb893

• cleanup and remove fast class transforms

  yonigozlan

  committed
  67d65f2

  Copy full SHA for 67d65f2

• update add fast image processor transformers cli

  yonigozlan

  committed
  d225448

  Copy full SHA for d225448

• cleanup docstring

  yonigozlan

  committed
  80c6824

  Copy full SHA for 80c6824

Commits on Jan 30, 2025

• Merge remote-tracking branch 'upstream/main' into improve-fast-image-processor-base

  yonigozlan

  committed
  b96adfa

  Copy full SHA for b96adfa

• uniformize pixtral fast and make _process_image explicit

  yonigozlan

  committed
  dbaacd1

  Copy full SHA for dbaacd1

• Merge remote-tracking branch 'upstream/main' into improve-fast-image-processor-base

  yonigozlan

  committed
  b660e9d

  Copy full SHA for b660e9d

• fix prepare image structure llava next/onevision

  yonigozlan

  committed
  b43ede1

  Copy full SHA for b43ede1

Commits on Feb 4, 2025

• Use typed kwargs instead of explicit args

  yonigozlan

  committed
  3b05cbd

  Copy full SHA for 3b05cbd

• nit fix import Unpack

  yonigozlan

  committed
  95db4a9

  Copy full SHA for 95db4a9

• Merge branch 'main' into improve-fast-image-processor-base

  yonigozlan

  authored
  d9e1fcd

  Copy full SHA for d9e1fcd

• clearly separate pops and gets in base preprocess. Use explicit typed kwargs

  yonigozlan

  committed
  6bd7a1b

  Copy full SHA for 6bd7a1b

• Merge branch 'main' into improve-fast-image-processor-base

  yonigozlan

  authored
  565e482

  Copy full SHA for 565e482

• make qwen2_vl preprocess arguments hashable

  yonigozlan

  committed
  f85c06f

  Copy full SHA for f85c06f

• Merge branch 'improve-fast-image-processor-base' of https://github.com/yonigozlan/transformers into improve-fast-image-processor-base

  yonigozlan

  committed
  1a7b0c4

  Copy full SHA for 1a7b0c4