Code

Add support for Apple's Depth-Pro#34583

Code

Merged

qubvel merged 170 commits intohuggingface:mainhuggingface/transformers:mainfrom geetu040:depth-progeetu040/transformers:depth-proCopy head branch name to clipboardFeb 10, 2025

+3,074-8

Commits

Commits on Nov 3, 2024

• implement config and model building blocks

  geetu040

  committed
  2986dc2

  Copy full SHA for 2986dc2

Commits on Nov 9, 2024

• refactor model architechture

  geetu040

  committed
  1728a2f

  Copy full SHA for 1728a2f

Commits on Nov 12, 2024

• update model outputs

  geetu040

  committed
  11ce50c

  Copy full SHA for 11ce50c

Commits on Nov 16, 2024

• update init param to include use_fov_model

  geetu040

  committed
  27e9593

  Copy full SHA for 27e9593

• update param name in config

  geetu040

  committed
  e74a7f5

  Copy full SHA for e74a7f5

• fix hidden_states and attentions outputs for fov

  geetu040

  committed
  8c2460b

  Copy full SHA for 8c2460b

• sort config

  geetu040

  committed
  55f6ed3

  Copy full SHA for 55f6ed3

• complete minor todos

  geetu040

  committed
  b25dffb

  Copy full SHA for b25dffb

• update patching

  geetu040

  committed
  c225deb

  Copy full SHA for c225deb

• update config for encoder

  geetu040

  committed
  176932d

  Copy full SHA for 176932d

• fix config

  geetu040

  committed
  dcec522

  Copy full SHA for dcec522

• use correct defaults in config

  geetu040

  committed
  0384d2f

  Copy full SHA for 0384d2f

Commits on Nov 17, 2024

• update merge for compatibility with different image size

  geetu040

  committed
  85e4f86

  Copy full SHA for 85e4f86

Commits on Nov 21, 2024

• restructure encoder for custom configuration

  geetu040

  committed
  00e4aa3

  Copy full SHA for 00e4aa3

• make fov model compatible with custom config

  geetu040

  committed
  6be242c

  Copy full SHA for 6be242c

• replace word "decoder" with "fusion"

  geetu040

  committed
  0189108

  Copy full SHA for 0189108

Commits on Nov 24, 2024

• weight conversion script

  geetu040

  committed
  7614e1a

  Copy full SHA for 7614e1a

Commits on Nov 25, 2024

• fix fov squeeze

  geetu040

  committed
  7d323ce

  Copy full SHA for 7d323ce

• update conversion script (without test)

  geetu040

  committed
  6aaa59e

  Copy full SHA for 6aaa59e

• upload ruff image processing

  geetu040

  committed
  263b773

  Copy full SHA for 263b773

Commits on Nov 26, 2024

• create fast image processing

  geetu040

  committed
  17e5487

  Copy full SHA for 17e5487

• use torch interpolation for image processing

  geetu040

  committed
  a8dd704

  Copy full SHA for a8dd704

• complete post_process_depth_estimation

  geetu040

  committed
  261bbaf

  Copy full SHA for 261bbaf

• config: fix imports and sort args

  geetu040

  committed
  a4b3556

  Copy full SHA for a4b3556

• apply inference in weight conversion

  geetu040

  committed
  f13c632

  Copy full SHA for f13c632

Commits on Nov 27, 2024

• use mllama script instead for weight conversion

  geetu040

  committed
  387ddd8

  Copy full SHA for 387ddd8

• clean weight conversion script

  geetu040

  committed
  9b67f9d

  Copy full SHA for 9b67f9d

• add depth-pro status in other files

  geetu040

  committed
  617c872

  Copy full SHA for 617c872

• fill docstring in config

  geetu040

  committed
  6e1c512

  Copy full SHA for 6e1c512

• formatting

  geetu040

  committed
  12ee607

  Copy full SHA for 12ee607

• more formatting

  geetu040

  committed
  d0a8733

  Copy full SHA for d0a8733

• formatting with ruff

  geetu040

  committed
  e6b385a

  Copy full SHA for e6b385a

• formatting with style

  geetu040

  committed
  267e50f

  Copy full SHA for 267e50f

• fix copied classes

  geetu040

  committed
  a1ec997

  Copy full SHA for a1ec997

• add examples; update weight convert script

  geetu040

  committed
  3c656f2

  Copy full SHA for 3c656f2

Commits on Nov 29, 2024

• fix using check_table.py and isort

  geetu040

  committed
  f6f6d3d

  Copy full SHA for f6f6d3d

• fix config docstring

  geetu040

  committed
  b4575d0

  Copy full SHA for b4575d0

• add depth pro to sdpa docs

  geetu040

  committed
  c8d8a9e

  Copy full SHA for c8d8a9e

• undo unintentional changes in configuration_gemma.py

  geetu040

  authored
  77873de

  Copy full SHA for 77873de

Commits on Nov 30, 2024

• minor fixes

  geetu040

  committed
  5f2378d

  Copy full SHA for 5f2378d

• test image processing

  geetu040

  committed
  d51d0b1

  Copy full SHA for d51d0b1

Commits on Dec 2, 2024

• fixes and tests

  geetu040

  committed
  082b055

  Copy full SHA for 082b055

• more fixes

  geetu040

  committed
  16a3917

  Copy full SHA for 16a3917

Commits on Dec 3, 2024

• use output states from image_encoder instead

  geetu040

  committed
  2408ec5

  Copy full SHA for 2408ec5

Commits on Dec 4, 2024

• Revert "use output states from image_encoder instead"

  Show description for be0c2a3
  geetu040

  committed
  be0c2a3

  Copy full SHA for be0c2a3

• make embeddings dynamic

  geetu040

  committed
  efed39f

  Copy full SHA for efed39f

• reshape output hidden states and attentions as part of computation graph

  geetu040

  committed
  c3b14fb

  Copy full SHA for c3b14fb

• fix ruff formating

  geetu040

  committed
  7cf2485

  Copy full SHA for 7cf2485

• fix docstring failure

  geetu040

  committed
  0aa451d

  Copy full SHA for 0aa451d

• use num_fov_head_layers in tests

  geetu040

  committed
  160afbf

  Copy full SHA for 160afbf

• update doc

  geetu040

  committed
  9d2be26

  Copy full SHA for 9d2be26

• check consistency with config

  geetu040

  committed
  e208459

  Copy full SHA for e208459

• ruff formatting

  geetu040

  committed
  0415722

  Copy full SHA for 0415722

• merge branch main

  geetu040

  committed
  402eedf

  Copy full SHA for 402eedf

Commits on Dec 5, 2024

• update test case

  geetu040

  committed
  f4e7404

  Copy full SHA for f4e7404

• fix ruff formatting

  geetu040

  committed
  2c1cc10

  Copy full SHA for 2c1cc10

• Merge branch 'main' into depth-pro

  geetu040

  committed
  4d94396

  Copy full SHA for 4d94396

Commits on Dec 6, 2024

• add tests for fov

  geetu040

  committed
  871b80d

  Copy full SHA for 871b80d

• use interpolation in postprocess

  geetu040

  committed
  0ff0655

  Copy full SHA for 0ff0655

• run and fix slow tests locally

  geetu040

  committed
  befa6cd

  Copy full SHA for befa6cd

• Merge branch 'main' into depth-pro

  geetu040

  committed
  db16fe6

  Copy full SHA for db16fe6

Commits on Dec 12, 2024

• use scaled_images_features for image and fov encoder

  geetu040

  committed
  99ac5e8

  Copy full SHA for 99ac5e8

• return fused_hidden_states in fusion stage

  geetu040

  committed
  ebb62dd

  Copy full SHA for ebb62dd

• fix example

  geetu040

  committed
  46c88e8

  Copy full SHA for 46c88e8

• fix ruff

  geetu040

  committed
  2431358

  Copy full SHA for 2431358

• Merge branch 'main' into depth-pro

  geetu040

  committed
  fd38841

  Copy full SHA for fd38841

Commits on Dec 21, 2024

• fix copyright license for all files

  geetu040

  committed
  d9d3a49

  Copy full SHA for d9d3a49

• add __all__ for each file

  geetu040

  committed
  8f4c61f

  Copy full SHA for 8f4c61f

• minor fixes

  Show description for 8960535
  geetu040

  committed
  8960535

  Copy full SHA for 8960535

• return list in post_process_depth_estimation

  geetu040

  committed
  1ac1b84

  Copy full SHA for 1ac1b84

• minor fixes

  Show description for 27bff69
  geetu040

  committed
  27bff69

  Copy full SHA for 27bff69

• fix "ruff check"

  geetu040

  committed
  a69b5af

  Copy full SHA for a69b5af

• update upsample and projection

  geetu040

  committed
  365a71d

  Copy full SHA for 365a71d

Commits on Dec 24, 2024

• major changes: (image size and merge optimization)

  Show description for c009468
  geetu040

  committed
  c009468

  Copy full SHA for c009468

• Merge branch 'main' into depth-pro

  geetu040

  committed
  7bed369

  Copy full SHA for 7bed369

• fix push_to_hub option in weights conversion

  geetu040

  committed
  1563f06

  Copy full SHA for 1563f06

• remove image_size in weights conversion

  geetu040

  committed
  e194ae4

  Copy full SHA for e194ae4

Commits on Jan 14, 2025

• major changes in the architecture

  Show description for a4889f2
  geetu040

  committed
  a4889f2

  Copy full SHA for a4889f2

• Merge branch "main"

  geetu040

  committed
  be5087b

  Copy full SHA for be5087b

• placeholder for unused config attributes

  geetu040

  committed
  9e09a6f

  Copy full SHA for 9e09a6f

• improve docs amid review

  geetu040

  committed
  bf159b2

  Copy full SHA for bf159b2

• minor change in docs

  geetu040

  committed
  fb41687

  Copy full SHA for fb41687

Commits on Jan 15, 2025

• further optimize merge

  geetu040

  committed
  7fbb53e

  Copy full SHA for 7fbb53e

• fix formatting

  geetu040

  committed
  558836c

  Copy full SHA for 558836c

Commits on Jan 24, 2025

• remove unused patch/batch convertion functions

  geetu040

  committed
  ed77f78

  Copy full SHA for ed77f78

• use original F.interpolate

  geetu040

  committed
  5bc4b31

  Copy full SHA for 5bc4b31

• improve function naming

  geetu040

  committed
  628ff09

  Copy full SHA for 628ff09

• minor chages

  Show description for e2996b6
  geetu040

  committed
  e2996b6

  Copy full SHA for e2996b6

• rearchitect upsample block for improved modularity

  geetu040

  committed
  8cb5c7a

  Copy full SHA for 8cb5c7a

• update upsample keys in weight conversion

  geetu040

  committed
  1ba3a4a

  Copy full SHA for 1ba3a4a

Commits on Jan 25, 2025

• improve padding in merge_patches

  geetu040

  committed
  83706b8

  Copy full SHA for 83706b8

• use double-loop for merge

  geetu040

  committed
  004cdc2

  Copy full SHA for 004cdc2

• update comments

  geetu040

  committed
  922b3de

  Copy full SHA for 922b3de

• create feature_extractor, reduce some forward code

  geetu040

  committed
  0f01b08

  Copy full SHA for 0f01b08

Commits on Jan 26, 2025

• introduce config.use_mask_token in dinov2

  geetu040

  committed
  4d871a7

  Copy full SHA for 4d871a7

• minor fixes

  geetu040

  committed
  85f7e3a

  Copy full SHA for 85f7e3a

• minor fixes for onnx

  geetu040

  committed
  c0127d7

  Copy full SHA for c0127d7

• update __init__ to latest format

  geetu040

  committed
  1898459

  Copy full SHA for 1898459

• remove DepthProConfig.to_dict()

  geetu040

  committed
  bcf1bf3

  Copy full SHA for bcf1bf3

• major changes in backbone

  geetu040

  committed
  09bffc3

  Copy full SHA for 09bffc3

• Merge branch 'main' into depth-pro

  geetu040

  committed
  0936897

  Copy full SHA for 0936897

• update config in weight conversion

  geetu040

  committed
  c26dc99

  Copy full SHA for c26dc99

• formatting

  geetu040

  committed
  5fb0bb7

  Copy full SHA for 5fb0bb7

• converted model is fp32

  geetu040

  committed
  d741890

  Copy full SHA for d741890

Commits on Jan 28, 2025

• improve naming and docs for feature_extractor->reconstruct_feature_maps

  geetu040

  committed
  03f137d

  Copy full SHA for 03f137d

• minor fixes; amid review

  geetu040

  committed
  2b8ee8f

  Copy full SHA for 2b8ee8f

• create intermediate vars in func call

  geetu040

  committed
  774617a

  Copy full SHA for 774617a

• use torch.testing.assert_close

  geetu040

  committed
  b6d15ff

  Copy full SHA for b6d15ff

• use ModuleList instead of Sequential and ModuleDict

  geetu040

  committed
  425d63e

  Copy full SHA for 425d63e

• update docs

  geetu040

  committed
  f415ee6

  Copy full SHA for f415ee6

• Merge branch 'main' into depth-pro

  geetu040

  committed
  2777305

  Copy full SHA for 2777305

Commits on Jan 30, 2025

• include fov in integraiton tests

  geetu040

  committed
  1a2dd3a

  Copy full SHA for 1a2dd3a

• update docs

  geetu040

  committed
  4cfebae

  Copy full SHA for 4cfebae

• improve initialization of convolution layers

  geetu040

  committed
  9062767

  Copy full SHA for 9062767

• fix unused fov keys

  geetu040

  committed
  fcba6bd

  Copy full SHA for fcba6bd

• update tests

  geetu040

  committed
  56cd570

  Copy full SHA for 56cd570

• Merge branch 'main' into depth-pro

  geetu040

  committed
  e64d39a

  Copy full SHA for e64d39a

• ruff format

  geetu040

  committed
  26b1391

  Copy full SHA for 26b1391

Commits on Jan 31, 2025

• Merge branch 'main' into depth-pro

  geetu040

  committed
  8914549

  Copy full SHA for 8914549

• fix test, amid kaimming initialization

  geetu040

  committed
  01247f8

  Copy full SHA for 01247f8

• add depthpro to toctree

  geetu040

  committed
  0b7e77f

  Copy full SHA for 0b7e77f

• add residual layer to _no_split_modules

  geetu040

  committed
  20b277d

  Copy full SHA for 20b277d

Commits on Feb 1, 2025

• architecture rework

  geetu040

  committed
  ff0e408

  Copy full SHA for ff0e408

• Update src/transformers/models/depth_pro/image_processing_depth_pro.py

  Show description for 1522c53
  

  geetu040

  and

  qubvel

  authored
  1522c53

  Copy full SHA for 1522c53

• Update src/transformers/models/depth_pro/image_processing_depth_pro_fast.py

  Show description for 131817a
  

  geetu040

  and

  qubvel

  authored
  131817a

  Copy full SHA for 131817a

• update docs

  geetu040

  committed
  72a1f0c

  Copy full SHA for 72a1f0c

• improve merge_patches

  geetu040

  committed
  aed7e3d

  Copy full SHA for aed7e3d

• use flatten with fov_output

  geetu040

  committed
  405bee3

  Copy full SHA for 405bee3

• ruff formatting

  geetu040

  committed
  a8528da

  Copy full SHA for a8528da

• Merge branch 'main' into depth-pro

  geetu040

  committed
  aed655c

  Copy full SHA for aed655c

Commits on Feb 3, 2025

• update resources section in docs

  Show description for 31383e1
  

  geetu040

  and

  qubvel

  authored
  31383e1

  Copy full SHA for 31383e1

• fix typo "final_kernal_size"

  Show description for 641cb84
  

  geetu040

  and

  qubvel

  authored
  641cb84

  Copy full SHA for 641cb84

• fix output typehint for DepthProDepthEstimator

  Show description for 6af8a11
  

  geetu040

  and

  qubvel

  authored
  6af8a11

  Copy full SHA for 6af8a11

• residual operation in 2 steps

  Show description for abd5307
  

  geetu040

  and

  qubvel

  authored
  abd5307

  Copy full SHA for abd5307

• use image_size instead of global patch_size in interpolation

  geetu040

  committed
  8dc2751

  Copy full SHA for 8dc2751

• replace all Sequential with ModuleList

  geetu040

  committed
  2f88694

  Copy full SHA for 2f88694

• update fov

  geetu040

  committed
  208ee26

  Copy full SHA for 208ee26

• update heads

  geetu040

  committed
  bc63511

  Copy full SHA for bc63511

• fix and update conversion script for heads

  geetu040

  committed
  e33a531

  Copy full SHA for e33a531

• ruff formatting

  geetu040

  committed
  8c0e81a

  Copy full SHA for 8c0e81a

• remove float32 conversion

  geetu040

  committed
  524dda6

  Copy full SHA for 524dda6

• Merge branch 'main' into depth-pro

  geetu040

  committed
  029dd9d

  Copy full SHA for 029dd9d

Commits on Feb 4, 2025

• use "Fov" instead of "FOV" in class names

  geetu040

  committed
  a87d26a

  Copy full SHA for a87d26a

• use "Fov" instead of "FOV" in config docs

  geetu040

  committed
  5fccbff

  Copy full SHA for 5fccbff

• remove prune_heads

  geetu040

  committed
  24f1413

  Copy full SHA for 24f1413

• update fusion stage

  geetu040

  committed
  a3dab18

  Copy full SHA for a3dab18

• use device in examples

  geetu040

  committed
  48eb534

  Copy full SHA for 48eb534

• Merge branch 'main' into depth-pro

  geetu040

  committed
  39ea929

  Copy full SHA for 39ea929

Commits on Feb 5, 2025

• Merge branch 'main' into depth-pro

  geetu040

  committed
  26db9ec

  Copy full SHA for 26db9ec

• update processor

  geetu040

  committed
  ba37c91

  Copy full SHA for ba37c91

• ruff fixes

  geetu040

  committed
  949ecb9

  Copy full SHA for 949ecb9

• add do_rescale in image_processor_dict

  geetu040

  committed
  0e2861d

  Copy full SHA for 0e2861d

• skip test: test_fast_is_faster_than_slow

  geetu040

  committed
  a6efedb

  Copy full SHA for a6efedb

• ruff formatting

  geetu040

  committed
  4d8f927

  Copy full SHA for 4d8f927

• DepthProImageProcessorFast in other files

  geetu040

  committed
  dd8de27

  Copy full SHA for dd8de27

• Merge branch 'main' into depth-pro

  geetu040

  authored
  75215ed

  Copy full SHA for 75215ed

• Merge branch 'main' into depth-pro

  geetu040

  authored
  ffb3a82

  Copy full SHA for ffb3a82

• revert antialias removal

  geetu040

  committed
  5caa0bd

  Copy full SHA for 5caa0bd

• add antialias in BaseImageProcessorFast

  geetu040

  committed
  3ae1134

  Copy full SHA for 3ae1134

• Revert "revert antialias removal"

  Show description for 8372ad9
  geetu040

  committed
  8372ad9

  Copy full SHA for 8372ad9

• Revert "add antialias in BaseImageProcessorFast"

  Show description for 666f3b7
  geetu040

  committed
  666f3b7

  Copy full SHA for 666f3b7

• update processor for grouping and antialias

  geetu040

  committed
  41180e3

  Copy full SHA for 41180e3

• try test_fast_is_faster_than_slow without "skip" or "flanky"

  geetu040

  committed
  1265b12

  Copy full SHA for 1265b12

• Merge branch 'main' into depth-pro

  geetu040

  authored
  86c4604

  Copy full SHA for 86c4604

• update checkpoint

  geetu040

  committed
  4dc850f

  Copy full SHA for 4dc850f

• Merge branch 'main' into depth-pro

  geetu040

  authored
  b7f32b9

  Copy full SHA for b7f32b9

Commits on Feb 6, 2025

• update checkpoint

  geetu040

  committed
  592648c

  Copy full SHA for 592648c

• use @is_flanky for processor test

  geetu040

  committed
  162f141

  Copy full SHA for 162f141

• Merge branch 'main' into depth-pro

  geetu040

  authored
  3a62d63

  Copy full SHA for 3a62d63

Commits on Feb 7, 2025

• update checkpoint to "apple/DepthPro-hf"

  geetu040

  committed
  4b76239

  Copy full SHA for 4b76239