Merge branch 'main' into rotated-bboxes-transforms

Author: AntoineSimoulin

.github/workflows/build-cmake.yml

@@ -18,7 +18,7 @@ jobs:
             gpu-arch-type: cpu
           - runner: linux.g5.4xlarge.nvidia.gpu
             gpu-arch-type: cuda
-            gpu-arch-version: "11.8"
+            gpu-arch-version: "12.6"
       fail-fast: false
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     permissions:
@@ -66,7 +66,7 @@ jobs:
             gpu-arch-type: cpu
           - runner: windows.g5.4xlarge.nvidia.gpu
             gpu-arch-type: cuda
-            gpu-arch-version: "11.8"
+            gpu-arch-version: "12.6"
       fail-fast: false
     uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
     with:

.github/workflows/build_wheel_windows_arm64.yml

@@ -0,0 +1,54 @@
+name: Build Windows ARM64 Wheels
+
+on:
+  pull_request:
+    paths:
+      - .github/workflows/build_wheel_windows_arm64.yml
+  push:
+    branches:
+      - nightly
+      - release/*
+    tags:
+        # NOTE: Binary build pipelines should only get triggered on release candidate builds
+        # Release candidate tags look like: v1.11.0-rc1
+        - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
+  workflow_dispatch:
+
+permissions:
+  id-token: write
+  contents: read
+
+jobs:
+  generate-matrix:
+    uses: pytorch/test-infra/.github/workflows/generate_binary_build_matrix.yml@main
+    with:
+      package-type: wheel
+      os: windows-arm64
+      test-infra-repository: pytorch/test-infra
+      test-infra-ref: main
+      with-cuda: disable
+
+  build:
+    needs: generate-matrix
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - repository: pytorch/vision
+            smoke-test-script: test/smoke_test.py
+            pre-script: packaging/pre_build_script_arm64.sh
+            package-name: torchvision
+            architecture: "arm64"
+    name: ${{ matrix.repository }}
+    uses: pytorch/test-infra/.github/workflows/build_wheels_windows.yml@main
+    with:
+      repository: ${{ matrix.repository }}
+      ref: ""
+      test-infra-repository: pytorch/test-infra
+      test-infra-ref: main
+      pre-script: ${{ matrix.pre-script }}
+      build-matrix: ${{ needs.generate-matrix.outputs.matrix }}
+      package-name: ${{ matrix.package-name }}
+      smoke-test-script: ${{ matrix.smoke-test-script }}
+      trigger-event: ${{ github.event_name }}
+      architecture: ${{ matrix.architecture }}

.github/workflows/prototype-tests-linux-gpu.yml

@@ -21,7 +21,7 @@ jobs:
           - python-version: "3.9"
             runner: linux.g5.4xlarge.nvidia.gpu
             gpu-arch-type: cuda
-            gpu-arch-version: "11.8"
+            gpu-arch-version: "12.6"
       fail-fast: false
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     permissions:

.github/workflows/tests.yml

@@ -24,7 +24,7 @@ jobs:
           - python-version: 3.9
             runner: linux.g5.4xlarge.nvidia.gpu
             gpu-arch-type: cuda
-            gpu-arch-version: "11.8"
+            gpu-arch-version: "12.6"
       fail-fast: false
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     permissions:

CONTRIBUTING.md

@@ -38,13 +38,13 @@ instructions](https://pytorch.org/get-started/locally/). Note that the official
 instructions may ask you to install torchvision itself. If you are doing development
 on torchvision, you should not install prebuilt torchvision packages.
 
-**Optionally**, install `libpng` and `libjpeg-turbo` if you want to enable
+**Optionally**, install `libpng`, `libjpeg-turbo` and `libwebp` if you want to enable
 support for
-native encoding / decoding of PNG and JPEG formats in
+native encoding / decoding of PNG, JPEG and WebP formats in
 [torchvision.io](https://pytorch.org/vision/stable/io.html#image):
 
 ```bash
-conda install libpng libjpeg-turbo -c pytorch
+conda install libpng libjpeg-turbo libwebp -c pytorch
 ```
 
 Note: you can use the `TORCHVISION_INCLUDE` and `TORCHVISION_LIBRARY`

packaging/pre_build_script_arm64.sh

@@ -6,9 +6,6 @@ echo "Building vision dependencies and wheel started."
 export SRC_PATH="$GITHUB_WORKSPACE/$SRC_DIR"
 export CMAKE_BUILD_TYPE="$BUILD_TYPE"
 export VCVARSALL_PATH="$DEPENDENCIES_DIR/VSBuildTools/VC/Auxiliary/Build/vcvarsall.bat"
-export CONDA_PREFIX="$DEPENDENCIES_DIR"
-export PATH="$PATH:$CONDA_PREFIX/Library/bin"
-export DISTUTILS_USE_SDK=1
 export TRIPLET_FILE="triplets/arm64-windows.cmake"
 export PYTORCH_VERSION="$PYTORCH_VERSION"
 export CHANNEL="$CHANNEL"

references/detection/coco_utils.py

@@ -123,7 +123,7 @@ def convert_to_coco_api(ds):
     coco_ds = COCO()
     # annotation IDs need to start at 1, not 0, see torchvision issue #1530
     ann_id = 1
-    dataset = {"images": [], "categories": [], "annotations": []}
+    dataset = {"images": [], "categories": [], "annotations": [], "info": {}}
     categories = set()
     for img_idx in range(len(ds)):
         # find better way to get target

test/test_ops.py

@@ -1201,67 +1201,32 @@ def test_forward_scriptability(self):
         torch.jit.script(ops.DeformConv2d(in_channels=8, out_channels=8, kernel_size=3))
 
 
-@pytest.mark.parametrize("dtype", (torch.float16, torch.float32, torch.float64))
-@pytest.mark.parametrize("device", cpu_and_cuda())
-@pytest.mark.parametrize("requires_grad", (True, False))
-def test_deform_conv2d_opcheck(dtype, device, requires_grad):
-    batch_size, channels_in, height, width = 1, 6, 10, 10
-    kernel_size = (3, 3)
-    stride = (1, 1)
-    padding = (1, 1)
-    dilation = (1, 1)
-    groups = 2
-    out_channels = 4
-    out_h = (height + 2 * padding[0] - dilation[0] * (kernel_size[0] - 1) - 1) // stride[0] + 1
-    out_w = (width + 2 * padding[1] - dilation[1] * (kernel_size[1] - 1) - 1) // stride[1] + 1
-    x = torch.randn(batch_size, channels_in, height, width, dtype=dtype, device=device, requires_grad=requires_grad)
-    offset = torch.randn(
-        batch_size,
-        2 * kernel_size[0] * kernel_size[1],
-        out_h,
-        out_w,
-        dtype=dtype,
-        device=device,
-        requires_grad=requires_grad,
-    )
-    weight = torch.randn(
-        out_channels,
-        channels_in // groups,
-        kernel_size[0],
-        kernel_size[1],
-        dtype=dtype,
-        device=device,
-        requires_grad=requires_grad,
-    )
-    bias = torch.randn(out_channels, dtype=dtype, device=device, requires_grad=requires_grad)
-    use_mask = True
-    mask = torch.sigmoid(
-        torch.randn(
-            batch_size,
-            kernel_size[0] * kernel_size[1],
-            out_h,
-            out_w,
-            dtype=dtype,
-            device=device,
-            requires_grad=requires_grad,
-        )
-    )
-    kwargs = {
-        "offset": offset,
-        "weight": weight,
-        "bias": bias,
-        "stride_h": stride[0],
-        "stride_w": stride[1],
-        "pad_h": padding[0],
-        "pad_w": padding[1],
-        "dilation_h": dilation[0],
-        "dilation_w": dilation[1],
-        "groups": groups,
-        "offset_groups": 1,
-        "use_mask": use_mask,
-        "mask": mask,  # no modulation in this test
-    }
-    optests.opcheck(torch.ops.torchvision.deform_conv2d, args=(x,), kwargs=kwargs)
+# NS: Remove me once backward is implemented for MPS
+def xfail_if_mps(x):
+    mps_xfail_param = pytest.param("mps", marks=(pytest.mark.needs_mps, pytest.mark.xfail))
+    new_pytestmark = []
+    for mark in x.pytestmark:
+        if isinstance(mark, pytest.Mark) and mark.name == "parametrize":
+            if mark.args[0] == "device":
+                params = cpu_and_cuda() + (mps_xfail_param,)
+                new_pytestmark.append(pytest.mark.parametrize("device", params))
+                continue
+        new_pytestmark.append(mark)
+    x.__dict__["pytestmark"] = new_pytestmark
+    return x
+
+
+optests.generate_opcheck_tests(
+    testcase=TestDeformConv,
+    namespaces=["torchvision"],
+    failures_dict_path=os.path.join(os.path.dirname(__file__), "optests_failures_dict.json"),
+    # Skip tests due to unimplemented backward
+    additional_decorators={
+        "test_aot_dispatch_dynamic__test_forward": [xfail_if_mps],
+        "test_autograd_registration__test_forward": [xfail_if_mps],
+    },
+    test_utils=OPTESTS,
+)
 
 
 class TestFrozenBNT:

torchvision/tv_tensors/_image.py

@@ -43,7 +43,7 @@ def __new__(
 
         tensor = cls._to_tensor(data, dtype=dtype, device=device, requires_grad=requires_grad)
         if tensor.ndim < 2:
-            raise ValueError
+            raise ValueError(f"Tensor must be 2D or higher, got {tensor.ndim}D tensor.")
         elif tensor.ndim == 2:
             tensor = tensor.unsqueeze(0)