Merge branch 'main' into jafelfeaef

test/test_transforms_v2.py

@@ -5203,6 +5203,11 @@ def test_transform(self, make_input, dtype, device):
             transforms.LinearTransformation(*self._make_matrix_and_vector(input)),
             input,
             check_sample_input=self._sample_input_adapter,
+            # Compat check is failing on M1 with:
+            # AssertionError: Tensor-likes are not close!
+            # Mismatched elements: 1 / 561 (0.2%)
+            # See https://github.com/pytorch/vision/issues/8453
+            check_v1_compatibility=(sys.platform != "darwin"),
         )
 
     def test_transform_error(self):

torchvision/io/video.py

@@ -244,7 +244,7 @@ def read_video(
     Reads a video from a file, returning both the video frames and the audio frames
 
     Args:
-        filename (str): path to the video file
+        filename (str): path to the video file. If using the pyav backend, this can be whatever ``av.open`` accepts.
         start_pts (int if pts_unit = 'pts', float / Fraction if pts_unit = 'sec', optional):
             The start presentation time of the video
         end_pts (int if pts_unit = 'pts', float / Fraction if pts_unit = 'sec', optional):
@@ -267,10 +267,9 @@ def read_video(
 
     from torchvision import get_video_backend
 
-    if not os.path.exists(filename):
-        raise RuntimeError(f"File not found: {filename}")
-
     if get_video_backend() != "pyav":
+        if not os.path.exists(filename):
+            raise RuntimeError(f"File not found: {filename}")
         vframes, aframes, info = _video_opt._read_video(filename, start_pts, end_pts, pts_unit)
     else:
         _check_av_available()

torchvision/transforms/v2/functional/_geometry.py

@@ -194,7 +194,7 @@ def resize(
 # according to our benchmarks on eager, non-AVX CPUs should still prefer u8->f32->interpolate->u8 path for bilinear
 def _do_native_uint8_resize_on_cpu(interpolation: InterpolationMode) -> bool:
     if interpolation == InterpolationMode.BILINEAR:
-        if torch._dynamo.is_compiling():
+        if torch.compiler.is_compiling():
             return True
         else:
             return "AVX2" in torch.backends.cpu.get_cpu_capability()
@@ -525,7 +525,7 @@ def _get_inverse_affine_matrix(
 
 
 def _compute_affine_output_size(matrix: List[float], w: int, h: int) -> Tuple[int, int]:
-    if torch._dynamo.is_compiling() and not torch.jit.is_scripting():
+    if torch.compiler.is_compiling() and not torch.jit.is_scripting():
         return _compute_affine_output_size_python(matrix, w, h)
     else:
         return _compute_affine_output_size_tensor(matrix, w, h)