GitLab

* add prototype features

* add some JIT tests

* refactor input data handling

* refactor tests

* cleanup tests

* add BoundingBox feature

* mypy

* xfail torchscript tests for now

* cleanup

* fix imports

* add tests for builtin prototype datasets

* fix caltech101

* fix emnist

* fix mnist and variants

* add iopath as test requirement

* fix MNIST warning

* fix qmnist data generation

* fix cifar data generation

* add tests for imagenet

* cleanup

* Moving builder to the bottom to use proper typing.

* Renaming weights.

* Adding quantizated inception builder.

* Correct meta info.

* Fix linter.

* Removing init_weights to avoid exposing it on the class.

* modify processPacket to support fast seek

* add fastSeek to ProcessPacket decoder definition

* add fastseek flag to DecoderParametersStruct

* add fastseek flag to the process packet call

* no default params in C++ implementation

* enable flag in C++ implementation

* make order of parameters more normal

* register new seek with python api

* [somewhat broken] test suite for keyframes using pyav

* revert " changes

* add type annotations to init

* Adding tests

* linter

* Flake doesn't show up :|

* Change from unitest to pytest syntax

* add return type
Co-authored-by: Prabhat Roy <prabhatroy@fb.com>

* Reordering the builders to use proper typing.

* Adding additional meta-data on existing quantized models.

* Fixing meta on unquantized model.

* Adding quantized googlenet builder.

* undo inception move.

* Adding recipe information.

* Aligning exception with all other models.

* Adding prototype preprocessing on video references.

* Adding the rest of model builders on faster_rcnn.

torchvision compatible with torch 1.9.1 was missing in table of supported versions.

* Fix TORCH_HOME environment variable in docs

* Update docs with torch.hub

Updated the reference for loading a model to `torch.hub`.
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* Adding prototype preprocessing on segmentation references.

* Adding prototype preprocessing on video references.

* Port semeion dataset

* Update torchvision/prototype/datasets/_builtin/semeion.py
Co-authored-by: Philip Meier <github.pmeier@posteo.de>

* explicitly convert the image array to torch.uint8

* explicitly convert the image array to torch.uint8
Co-authored-by: Philip Meier <github.pmeier@posteo.de>
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* use `store_test_results` for prototype tests

* add failing test

* Revert "add failing test"

This reverts commit ebe2b37cb50461031a94131fe4bd3fae71b48e34.

* [references] Adding gradient clipping

* ufmt formatting

* remove apex code

* resolve naming issue

Fix for https://github.com/pytorch/vision/issues/4832

The unauthenticated git protocol on port 9418 is no longer supported.
Please see https://github.blog/2021-09-01-improving-git-protocol-security-github/ for more information.

* Update typing

* Fix bug

* Unblock mypy

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

* Adding multi-weight support to Quantized ResNet.

* Update references script to support testing quantized models with the new API.

* Handle quantized models correctly in ref script.

* Fixing references for quantization.

* Update training references from legacy models.

* Refactoring to share common parts.

* Expose misc ops at package level

* Adding documentation to the ops exposed

* Moving original builder at the bottom of the page to use proper typing.

* Adding multiweight support to inception.

* Update doc.

* change range into uint8

* Same typo in TrivialAugmentWide and RandAugment
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Prabhat Roy <prabhatroy@fb.com>

* Change enum name for weights contributed by community.

* Adding multiweight support to squeezenet.

* improve download nightly pytorch

* update unitest cuda to 11.3

* add nvjpeg in cuda 11.3

* avoid catching exception

* restore to cu111

* Update install.sh

* use conda MatchSpec string

* update linux

* for linux

* Update install.sh
Co-authored-by: Nikita Shulga <nshulga@fb.com>

* Move model builder at the bottom of the file, so we can use proper typing.

* Adding GoogLeNet with multi-weight support.

* Simplify expression.

* add tests for prototype datasets API

* add pytest-mock as test dependency

* remove unnecessary fixtures

* Apply suggestions from code review
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* address first batch of review comments

* remove dynamic class generation

* aggregate similar tests

* split immutable tests once again

* removed unused fixture

* remove re.escape usages

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

* Add logging to torchvision ops

* Hack to make torchscript work

* Bugfix

* Bugfix

* Lint

* mypy... let's silence it

* Fighting with mymy

* One more try

Summary:
This diff enables ImageNet to be usable from within torchvision.

It implements a different codepath for some families of datasets that can be sharded into multiple pieces, and uses it if within fbcode.

For now all previously supported datasets are also supported (MNIST, CIFAR and FashionMNIST), in addition to ImageNet.

Pull Request resolved: https://github.com/pytorch/vision/pull/4776

Reviewed By: datumbox

Differential Revision: D31929684

fbshipit-source-id: 58f9fe2ed6ce731a6b43a38be912983247eff562
(cherry picked from commit b43fead8ea5164e37637ade0e80477e94f993365)

* Update readme for new torch

* Fix typo

It's FBcode, not FBCode

* Adding multiweight support for shufflenetv2 prototype models

* Revert "Adding multiweight support for shufflenetv2 prototype models"

This reverts commit 31fadbee7d1a65cd73ae43dfd4ac6e97e7ca7b01.

* Adding multiweight support for shufflenetv2 prototype models

* Revert "Adding multiweight support for shufflenetv2 prototype models"

This reverts commit 4e3d900f796c1e3e667312087e77956ca4a4c017.

* Adding multiweight support for shufflenetv2 prototype models

* addressing PR comment