GitLab

According to the benchmark link https://github.com/pytorch/vision/issues/1311#issuecomment-781329339

, for GPU the threshold should be higher (which is why detectron2 used 40k).
This PR changes the threshold to be device dependent.
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* Add bias parameter to ConvNormActivation

* Update torchvision/ops/misc.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

Co-authored-by: Prabhat Roy <prabhatroy@fb.com>

making torchvision ops leaf nodes by default

* improve COCO prototype

* test 2017 annotations

* add option to include captions

* fix categories and add tests

* cleanup

* add correct image size to bounding boxes

* fix annotation collation

* appease mypy

* add benchmark

* always use image as reference

* another refactor

* add support for segmentations

* add support for segmentations

* fix CI dependencies

* Change the `default` weights mechanism to sue Enum aliases.

* Change `get_weights` to work with full Enum names and make it public.

* Applying improvements from code review.

* [DIRTY] cleanup CI config

* remove pip install command

* fix syntax

* fix jobs

* fix syntax

* add support for editable install

* sync

* fix sudo apt install

* fix editable install

* sync

* try pip without user install

* [DIRTY] address review comments

* [DIRTY] try use dynamic name

* switch logic

* address remaining comments

* cleanup

* more cleanup

* fix enabling prototype tests

* Update .circleci/config.yml.in
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* linebreak
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* Fix bug on autocontrast when `min==max`

* Adding PIL vs TorchVision test for min==max

* Rename classes Weights => WeightsEnum and WeightEntry => Weights.

* Make enum values follow the naming convention `_V1`, `_V2` etc

* Cleanup the Enum class naming conventions.

* Add a test to check naming conventions.

Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Update functional_tensor.py

To resolve issue #4818
Add assert function and logic after checking bound of image

* Update functional_tensor.py

* Update test

* Update test_functional_tensor.py

* Update test_functional_tensor.py

* Update test_functional_tensor.py

* Update test_functional_tensor.py

* Update test_functional_tensor.py

* Update test_functional_tensor.py

* Update test_functional_tensor.py

* Update test_functional_tensor.py

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

* [vit] Adding ViT to torchvision/models

* adding pre-logits layer + resolving comments

* Fix the model attribute bug

* Change version to arch

* fix failing unittests

* remove useless prints

* reduce input size to fix unittests

* Increase windows-cpu executor to 2xlarge

* Use `batch_first=True` and remove classifier

* Change resource_class back to xlarge

* Remove vit_h_14

* Remove vit_h_14 from __all__

* Move vision_transformer.py into prototype

* Fix formatting issue

* remove arch in builder

* Fix type err in model builder

* address comments and trigger unittests

* remove the prototype import in torchvision.models

* Adding vit back to models to trigger CircleCI test

* fix test_jit_forward_backward

* Move all to prototype.

* Adopt new helper methods and fix prototype tests.

* Remove unused import.
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Vasilis Vryniotis <vvryniotis@fb.com>

* enable caching of model weights for prototype CI

* syntax

* syntax

* make cache dir dynamic

* increase verbosity

* fix

* use larget CI machine

* revert debug output

* [DEBUG] test env var usage in save_cache

* retry

* use checksum for caching

* remove env vars because expansion is not working

* syntax

* cleanup

* base caching on model-urls

* relax regex

* cleanup skips

* cleanup

* fix skipping logic

* improve step name

* benchmark without caching

* benchmark with external download

* debug

* fix manual download location

* debug again

* download weights in the background

* try parallel download

* add missing import

* use correct decoractor

* up resource_class

* fix wording

* enable stdout passthrough to see download during test

* remove linebreak

* move checkout up

* cleanup

* debug failing test

* temp fix

* fix

* cleanup

* fix regex

* remove explicit install of numpy

* support amp training for segmention models

* fix lint
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Updated the link for densenet recipe.

* Set default value of `num_classes` and `num_keypoints` to `None`

* Provide helper methods for parameter checks to reduce duplicate code.

* Throw errors on silent config overwrites from weight meta-data and legacy builders.

* Changing order of arguments + fixing mypy.

* Make the builders fully BC.

* Add "default" weights support that returns always the best weights.

Otherwise, it `setuptools.py`  will try to install latest, which is not compatible with Python runtime older than 3.8

Fixes https://github.com/pytorch/vision/issues/4985

* style: Fixed last missing typing annotation

* style: Fixed typing

* style: Fixed remaining typing annotations

* style: Fixed typing

* style: Fixed typing

* refactor: Removed unused import

* Update torchvision/datasets/video_utils.py
Co-authored-by: Philip Meier <github.pmeier@posteo.de>
Co-authored-by: Prabhat Roy <prabhatroy@fb.com>

* don't close file handles in protoype dataset tests

* remove file closing from decoder
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* change to stable sort in nms implementations

Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* add third party dependencies to prototype datasets

* add test for Dataset.to_datapipe

* add missing annotation

* Add multi-type support on get_weight.

* Fix bug on method call.

* Adding logging suffix for QAT.

* Update wide_resnet101 accuracy values and path to improved model

* Fixed github link

* add FloReader datapipe

* add NumericBinaryReader

* revert unrelated change

* cleanup

* cleanup

* add comment for byte reversal

* use numpy after all

* appease mypy

* use .astype() with copy=False

* add docstring and cleanuo

* reuse current _read_flo and revert MNIST changes

* cleanup

* revert demonstration

* refactor

* cleanup

* add support for mutable memory

* add test

* add comments

* catch more exceptions

* fix mypy

* fix variable names

* hardcode flow sizes in test

* add fix dtype docstring

* expand comment on different reading modes

* add comment about files in update mode

* add tests for fromfile

* cleanup

* cleanup

Co-authored-by: Philip Meier <github.pmeier@posteo.de>

Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* style: Added typing annotations

* style: Fixed lint

* style: Fixed typing

* chore: Updated mypy.ini

* style: Fixed typing

* chore: Updated mypy.ini

* style: Fixed typing compatibility with jit

* style: Fixed typing

* style: Fixed typing

* style: Fixed missing import

* style: Fixed typing of __iter__

* style: Fixed typing

* style: Fixed lint

* style: Finished typing

* style: ufmt the file

* style: Removed unnecessary typing

* style: Fixed typing of iterator
Co-authored-by: Philip Meier <github.pmeier@posteo.de>
Co-authored-by: Prabhat Roy <prabhatroy@fb.com>

* return features instead of vanilla tensors from prototype datasets

* fix tests

* remove inplace

* add explanation for __init_subclass__

* fix label for test split

* relax test

* remove pixels