GitLab

[proto] Consistent supported/unsupported types handling in LinearTransformation, other perf comments (#6498)

* WIP

* [proto] unformal supported/unsupported types handling in LinearTransformation, other perf comments

* Type fixes and other minor stuff

* Fix pass-through and supported types of Normalize

* update error message on kernel

* Fix linter.

* Fix the tests.

* Update type.

* Update type.

* Remove unnecessary tests for bboxes and masks.

[proto] Use the proper `_transformed_types` in all Transforms and eliminate unnecessary dispatching (#6494)

* Update types in deprecated transforms.

* Update types in type conversion transforms.

* Fixing types in meta transforms.

* More changes on type conversion.

* Bug fix.

* Fix types

* Remove unnecessary conversions.

* Remove unnecessary import.

* Fixing tests

* Remove copy support from `to_image_tensor`

* restore test param

* Fix further tests

* [proto] Updated tests for CopyPaste on OneHotLabel

* Fixing test error

* Fixes unexpected behaviour with Transform._transformed_types and torch.Tensor

* Make code consistent to has_any, has_all implementation

* Fixed failing flake8 check

* add deprecated color conversion functionals

* allow grayscale tensor inputs in rgb_to_grayscale

* add cloning to tensor no-op

* improve todo comment

* [skip ci]

* use legacy kernels

use legacy kernels

* fix import

* remove duplicate code

* remove duplicate check

* Fix out-of-bounds read in decode_png

* Fix lint

* Use size_t instead of int to fix sign-compare check

* Add a unit-test to decode currupted png file

* Fix linter

* Update torchvision/csrc/io/image/cpu/decode_png.cpp

* Replace png_error with TORCH_CHECK
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Update error message

* Update error message
Co-authored-by: vfdev <vfdev.5@gmail.com>

* Fix linter
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: vfdev <vfdev.5@gmail.com>

* WIP

* [proto] Added SimpleCopyPaste transform

* Refactored and cleaned the implementation and added tests

* Fixing code

* Fixed code formatting issue

* Minor updates

* Fixed merge issue
Co-authored-by: Philip Meier <github.pmeier@posteo.de>

* add dispatcher for erase image kernels

* simplify RandomErasing

* replace query_image with query_image_dimensions

* use value sentinel and fix test for RandomErasing

* image_dimensions -> chw

* WIP

* Fixed dtype correction and tests

* Removed PIL Image support and output always Tensor

* [proto] Fixed to_image_tensor and to_image_pil

* Fixed failing test

* S3D initial commit

* add model builder code and docstrings

* change classifier submodule, populate weights enum

* fix change of block args from List[List[int]] to ints

* add VideoClassification to transforms

* edit weights url for testing, add s3d to models.video init

* norm_layer changes

* norm_layer and args fix

* Overwrite default dropout

* Remove docs from internal submodules.

* Fix tests

* Adding documentation.

* Link doc from main models.rst

* Fix min_temporal_size

* Adding crop/resize parameters in references script

* Release weights.

* Refactor dropout.

* Adding the weights table in the doc
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Vasilis Vryniotis <vvryniotis@fb.com>

* port `FixedSizeCrop` from detection references to prototype transforms

* mypy

* [skip ci] call invalid boxes and corresponding masks and labels

* cherry-pick missing functions from #6401

* fix feature wrapping

* add test

* mypy

* add input type restrictions

* add test for _get_params

* fix input checks

* Added ETH3D stereo dataset

* Small doc-reformating

* Removed assertions with no use, changed np conversion

* Added ETH3D stereo dataset

* Removed assertions with no use, changed np conversion

* rebased on main

* Revert "Removed assertions with no use, changed np conversion"

This reverts commit 1478a8c056bbed59b6f4a67f78a5cfac84cc9fda.

* Update to np.bool instead of np.bool_

* lint and mypy nit fix

* test nit

* expand has_any and has_all to also accept check callables

* add test and fix has_all

* add support for simple tensor images to CutMix, MixUp and RandomIoUCrop

* remove TODO

* remove pythonic syntax sugar

* simplify

* use concreate examples in test rather than abstract ones

* simplify further

* [proto] Ported RandomIoUCrop from detection refs

* Scope acceptable data types

* Added get_params test

* Added test__transform_empty_params

* Added support for OneHotLabel and tests

* Added tests for mask

* Updated error message

* Apply suggestions from code review
Co-authored-by: Philip Meier <github.pmeier@posteo.de>

* Added support for OHE masks and tests

* Ignored mypy error

* Fixed forward call on sample

* Added a todo
Co-authored-by: Philip Meier <github.pmeier@posteo.de>

* Added Falling Things datasets

* Renamed split to variant

* Update torchvision/datasets/_stereo_matching.py

Changed constant formatting
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* added SceneFLow variant datasets

* Changed split name to variant name

* removed trailing commented code line

* Added InStereo2k dataset

* Added Sintel Stereo dataset

* small refactor in tests

* Fixed doc formatting.

* candidate fix for FileNotFound on windows test

* Adressing comments

* Added Sintel Stereo dataset

* small refactor in tests

* Fixed doc formatting.

* candidate fix for FileNotFound on windows test

* Adressing comments

* rebased on main

* lint fix

* Added InStereo2k dataset

* added SceneFLow variant datasets

* Changed split name to variant name

* removed trailing commented code line

* Added Sintel Stereo dataset

* small refactor in tests

* Fixed doc formatting.

* candidate fix for FileNotFound on windows test

* Adressing comments

* Added Sintel Stereo dataset

* small refactor in tests

* Fixed doc formatting.

* candidate fix for FileNotFound on windows test

* Adressing comments

* rebased on main

* lint fix

* port `RandomShortestSize` from detection references to prototype transforms

* mypy

* add test

* fix TenCrop

* use dispatchers for RandomPhotometricDistort

* add convert_color_space dispatcher and use it in conversion transforms

* fix convert_color_space naming scheme

* add to_color_space method to Image feature

* remove TODO from BoundingBox.to_format()

* fix test

* fix imports

* fix passthrough

* remove apply_recursively in favor of pytree

* refactor BatchMultiCrop

* port ScaleJitter from detection reference to prototype transforms

* add test

* use MagicMock as sentinel

* disable download test for Places365

* add xfail test for broken download servers

* add more comments

* mark SBU download as failing
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* [proto] Another attemp to rewrite RandomCrop

* Fixed implementation issue and updated tests

* [proto] Compose keeps BC

* Compose -> Compose(Transform)

* Fix typing jit issue on RoIPool and RoIAlign

* Fix nit.

* Address code review comments.

* init submit

* fix typo

* support ufmt and mypy

* fix 2 unittest errors

* fix ufmt issue

* Apply suggestions from code review
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* unify codes

* fix meshgrid indexing

* fix a bug

* fix type check

* add type_annotation

* add slow model

* fix device issue

* fix ufmt issue

* add expect pickle file

* fix jit script issue

* fix type check

* keep consistent argument order

* add support for pretrained_window_size

* avoid code duplication

* a better code reuse

* update window_size argument

* make permute and flatten operations modular

* add PatchMergingV2

* modify expect.pkl

* use None as default argument value

* fix type check

* fix indent

* fix window_size (temporarily)

* remove "v2_" related prefix and add v2 builder

* remove v2 builder

* keep default value consistent with official repo

* deprecate dropout

* deprecate pretrained_window_size

* fix dynamic padding edge case

* remove unused imports

* remove doc modification

* Revert "deprecate dropout"

This reverts commit 8a13f932815ae25655c07430d52929f86b1ca479.

* Revert "fix dynamic padding edge case"

This reverts commit 1c7579cb1bd7bf2f0f94907f39bee6ed707a97a8.

* remove unused kwargs

* add downsample docs

* revert block default value

* revert argument order change

* explicitly specify start_dim

* add small and base variants

* add expect files and slow_models

* Add model weights and documentation for swin v2

* fix lint

* fix end of files line
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Joao Gomes <jdsgomes@fb.com>

* Extending to support MViTv2

* Fix docs, mypy and linter

* Refactor the relative positional code.

* Code refactoring.

* Rename vars.

* Update docs.

* Replace assert with exception.

* Updat docs.

* Minor refactoring.

* Remove the square input limitation.

* Moving methods around.

* Modify the shortcut in the attention layer.

* Add ported weights.

* Introduce a `residual_cls` config on the attention layer.

* Make the patch_embed kernel/padding/stride configurable.

* Apply changes from code-review.

* Remove stale todo.

* [proto] Fixed issue with `F.pad` from RandomZoomOut

* Fixed failing tests

* Fixed wrong type hint

* Fixed fill=None in pad_image_pil

* Try to support fill=None in functional

* Code formatting

* [proto] Fix for handling Numpy arrays by Transform

* transformed_types -> _transformed_types

* Added Image.normalize and fixed Normalize._transform

* Updated code for normalize, removed Image.normalize

* Make test precision stricter for Classification

* Update classification threshold.

* Update quantized classification threshold.

* added SceneFLow variant datasets

* Changed split name to variant name

* removed trailing commented code line

* Broken down PR(#6269). Added an additional dataset

* Removed some types. Store None instead of "". Merged test util functions.

* minor mypy fixes. minor doc fixes

* reformated docstring

* Added additional line-skips

* Fix bug in calculating cIoU for unequal sizes

* Remove comment

* what the epsilon?

* Fixing DIoU

* Optimization by Francisco.

* Fix the expected values on CompleteBoxIoU

* Apply suggestions from code review
Co-authored-by: Abhijit Deo <72816663+abhi-glitchhg@users.noreply.github.com>

* Adding cartesian product test.

* remove static
Co-authored-by: Vasilis Vryniotis <vvryniotis@fb.com>
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Abhijit Deo <72816663+abhi-glitchhg@users.noreply.github.com>

* cleaning up box decoding

* minor nits

* cleanup for box encoding also addded.