Merge branch 'main' into Add-Allowed-Workflows-for-Retry-Bot

.github/scripts/setup-env.sh

@@ -14,23 +14,26 @@ case $(uname) in
   Darwin)
     OS_TYPE=macos
     ;;
+  MSYS*)
+    OS_TYPE=windows
+    ;;
   *)
     echo "Unknown OS type:" $(uname)
     exit 1
     ;;
 esac
 
-echo '::group::Uninstall system JPEG libraries on macOS'
-# The x86 macOS runners, e.g. the GitHub Actions native "macos-12" runner, has some JPEG libraries installed by default
-# that interfere with our build. We uninstall them here and use the one from conda below.
 if [[ "${OS_TYPE}" == "macos" && $(uname -m) == x86_64 ]]; then
+  echo '::group::Uninstall system JPEG libraries on macOS'
+  # The x86 macOS runners, e.g. the GitHub Actions native "macos-12" runner, has some JPEG libraries installed by
+  # default that interfere with our build. We uninstall them here and use the one from conda below.
   JPEG_LIBS=$(brew list | grep jpeg)
   echo $JPEG_LIBS
   for lib in $JPEG_LIBS; do
     brew uninstall --ignore-dependencies --force $lib || true
   done
+  echo '::endgroup::'
 fi
-echo '::endgroup::'
 
 echo '::group::Create build environment'
 # See https://github.com/pytorch/vision/issues/7296 for ffmpeg
@@ -66,10 +69,24 @@ ltt install --progress-bar=off \
   torch
 
 if [[ $GPU_ARCH_TYPE == 'cuda' ]]; then
-  python3 -c "import torch; exit(not torch.cuda.is_available())"
+  python -c "import torch; exit(not torch.cuda.is_available())"
 fi
 echo '::endgroup::'
 
+if [[ "${OS_TYPE}" == "windows" ]]; then
+  echo '::group::Install third party dependencies prior to TorchVision install on Windows'
+  # `easy_install`, i.e. `python setup.py` has problems downloading the dependencies due to SSL.
+  # Thus, we install them upfront with `pip` to avoid that.
+  # Instead of fixing the SSL error, we can probably maintain this special case until we switch away from the deprecated
+  # `easy_install` anyway.
+  python setup.py egg_info
+  # The requires.txt cannot be used with `pip install -r` directly. The requirements are listed at the top and the
+  # optional dependencies come in non-standard syntax after a blank line. Thus, we just extract the header.
+  sed -e '/^$/,$d' *.egg-info/requires.txt > requirements.txt
+  pip install --progress-bar=off -r requirements.txt
+  echo '::endgroup::'
+fi
+
 echo '::group::Install TorchVision'
 python setup.py develop
 echo '::endgroup::'

.github/scripts/unittest.sh

@@ -4,15 +4,11 @@ set -euo pipefail
 
 ./.github/scripts/setup-env.sh
 
-# Prepare conda
-CONDA_PATH=$(which conda)
-eval "$(${CONDA_PATH} shell.bash hook)"
-conda activate ci
+# Activate conda environment
+eval "$($(which conda) shell.bash hook)" && conda deactivate && conda activate ci
 
 echo '::group::Install testing utilities'
 pip install --progress-bar=off pytest pytest-mock pytest-cov
 echo '::endgroup::'
 
-echo '::group::Run unittests'
 pytest --junit-xml="${RUNNER_TEST_RESULTS_DIR}/test-results.xml" -v --durations=25
-echo '::endgroup::'

.github/workflows/test-windows.yml

@@ -0,0 +1,51 @@
+name: Tests on Windows
+
+on:
+  pull_request:
+  push:
+    branches:
+      - nightly
+      - main
+      - release/*
+  workflow_dispatch:
+
+jobs:
+  unittests:
+    strategy:
+      matrix:
+        python-version:
+          - "3.8"
+          - "3.9"
+          - "3.10"
+          - "3.11"
+        runner: ["windows.4xlarge"]
+        gpu-arch-type: ["cpu"]
+        # FIXME: enable this as soon as nvjpeg is available on the Windows runner
+#        include:
+#          - python-version: "3.8"
+#            runner: windows.8xlarge.nvidia.gpu
+#            gpu-arch-type: cuda
+#            gpu-arch-version: "11.7"
+      fail-fast: false
+    uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
+    with:
+      repository: pytorch/vision
+      runner: ${{ matrix.runner }}
+      timeout: 120
+      script: |
+        set -euxo pipefail
+
+        export PYTHON_VERSION=${{ matrix.python-version }}
+        export GPU_ARCH_TYPE=${{ matrix.gpu-arch-type }}
+        export GPU_ARCH_VERSION=${{ matrix.gpu-arch-version }}
+        
+        # TODO: Port this to pytorch/test-infra/.github/workflows/windows_job.yml
+        export PATH="/c/Jenkins/Miniconda3/Scripts:${PATH}"
+        
+        if [[ $GPU_ARCH_TYPE == 'cuda' ]]; then
+          # TODO: This should be handled by the generic Windows job the same as its done by the generic Linux job
+          export CUDA_HOME="/c/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v${{ matrix.gpu-arch-version }}"
+          export CUDA_PATH="${CUDA_HOME}"
+        fi
+        
+        ./.github/scripts/unittest.sh

torchvision/models/video/mvit.py

@@ -765,9 +765,10 @@ def mvit_v1_b(*, weights: Optional[MViT_V1_B_Weights] = None, progress: bool = T
 @register_model()
 @handle_legacy_interface(weights=("pretrained", MViT_V2_S_Weights.KINETICS400_V1))
 def mvit_v2_s(*, weights: Optional[MViT_V2_S_Weights] = None, progress: bool = True, **kwargs: Any) -> MViT:
-    """
-    Constructs a small MViTV2 architecture from
-    `Multiscale Vision Transformers <https://arxiv.org/abs/2104.11227>`__.
+    """Constructs a small MViTV2 architecture from
+    `Multiscale Vision Transformers <https://arxiv.org/abs/2104.11227>`__ and
+    `MViTv2: Improved Multiscale Vision Transformers for Classification
+    and Detection <https://arxiv.org/abs/2112.01526>`__.
 
     .. betastatus:: video module
 
@@ -785,7 +786,7 @@ def mvit_v2_s(*, weights: Optional[MViT_V2_S_Weights] = None, progress: bool = T
             for more details about this class.
 
     .. autoclass:: torchvision.models.video.MViT_V2_S_Weights
-        :members:
+            :members:
     """
     weights = MViT_V2_S_Weights.verify(weights)