Merge branch 'main' of github.com:pytorch/vision into extra_decoders

Author: NicolasHug

.github/workflows/update-viablestrict.yml

@@ -27,6 +27,15 @@ All the datasets have almost similar API. They all have two common arguments:
 ``transform`` and  ``target_transform`` to transform the input and target respectively.
 You can also create your own datasets using the provided :ref:`base classes <base_classes_datasets>`.
 
+.. warning::
+
+    When a dataset object is created with ``download=True``, the files are first
+    downloaded and extracted in the root directory. This download logic is not
+    multi-process safe, so it may lead to conflicts / race conditions if it is
+    run within a distributed setting. In distributed mode, we recommend creating
+    a dummy dataset object to trigger the download logic *before* setting up
+    distributed mode.
+
 Image classification
 ~~~~~~~~~~~~~~~~~~~~
 

docs/source/datasets.rst

@@ -42,7 +42,7 @@
 IS_ROCM = (torch.version.hip is not None) and (ROCM_HOME is not None)
 BUILD_CUDA_SOURCES = (torch.cuda.is_available() and ((CUDA_HOME is not None) or IS_ROCM)) or FORCE_CUDA
 
-PACKAGE_NAME = "torchvision"
+package_name = os.getenv("TORCHVISION_PACKAGE_NAME", "torchvision")
 
 print("Torchvision build configuration:")
 print(f"{FORCE_CUDA = }")
@@ -98,7 +98,7 @@ def get_dist(pkgname):
         except DistributionNotFound:
             return None
 
-    pytorch_dep = "torch"
+    pytorch_dep = os.getenv("TORCH_PACKAGE_NAME", "torch")
     if os.getenv("PYTORCH_VERSION"):
         pytorch_dep += "==" + os.getenv("PYTORCH_VERSION")
 
@@ -127,7 +127,7 @@ def get_macros_and_flags():
             if NVCC_FLAGS is None:
                 nvcc_flags = []
             else:
-                nvcc_flags = nvcc_flags.split(" ")
+                nvcc_flags = NVCC_FLAGS.split(" ")
         extra_compile_args["nvcc"] = nvcc_flags
 
     if sys.platform == "win32":
@@ -561,7 +561,7 @@ def run(self):
     version, sha = get_version()
     write_version_file(version, sha)
 
-    print(f"Building wheel {PACKAGE_NAME}-{version}")
+    print(f"Building wheel {package_name}-{version}")
 
     with open("README.md") as f:
         readme = f.read()
@@ -573,7 +573,7 @@ def run(self):
     ]
 
     setup(
-        name=PACKAGE_NAME,
+        name=package_name,
         version=version,
         author="PyTorch Core Team",
         author_email="[email protected]",
@@ -583,7 +583,7 @@ def run(self):
         long_description_content_type="text/markdown",
         license="BSD",
         packages=find_packages(exclude=("test",)),
-        package_data={PACKAGE_NAME: ["*.dll", "*.dylib", "*.so", "prototype/datasets/_builtin/*.categories"]},
+        package_data={package_name: ["*.dll", "*.dylib", "*.so", "prototype/datasets/_builtin/*.categories"]},
         zip_safe=False,
         install_requires=get_requirements(),
         extras_require={

setup.py

@@ -1869,11 +1869,6 @@ def test_class_to_idx(self):
         with self.create_dataset() as (dataset, _):
             assert dataset.class_to_idx == class_to_idx
 
-    def test_images_download_preexisting(self):
-        with pytest.raises(RuntimeError):
-            with self.create_dataset({"download": True}):
-                pass
-
 
 class INaturalistTestCase(datasets_utils.ImageDatasetTestCase):
     DATASET_CLASS = datasets.INaturalist

test/test_datasets.py

@@ -6169,3 +6169,50 @@ def test_transform_sequence_len_error(self, quality):
     def test_transform_invalid_quality_error(self, quality):
         with pytest.raises(ValueError, match="quality must be an integer from 1 to 100"):
             transforms.JPEG(quality=quality)
+
+
+class TestUtils:
+    # TODO: Still need to test has_all, has_any, check_type and get_bouding_boxes
+    @pytest.mark.parametrize(
+        "make_input1", [make_image_tensor, make_image_pil, make_image, make_bounding_boxes, make_segmentation_mask]
+    )
+    @pytest.mark.parametrize(
+        "make_input2", [make_image_tensor, make_image_pil, make_image, make_bounding_boxes, make_segmentation_mask]
+    )
+    @pytest.mark.parametrize("query", [transforms.query_size, transforms.query_chw])
+    def test_query_size_and_query_chw(self, make_input1, make_input2, query):
+        size = (32, 64)
+        input1 = make_input1(size)
+        input2 = make_input2(size)
+
+        if query is transforms.query_chw and not any(
+            transforms.check_type(inpt, (is_pure_tensor, tv_tensors.Image, PIL.Image.Image, tv_tensors.Video))
+            for inpt in (input1, input2)
+        ):
+            return
+
+        expected = size if query is transforms.query_size else ((3,) + size)
+        assert query([input1, input2]) == expected
+
+    @pytest.mark.parametrize(
+        "make_input1", [make_image_tensor, make_image_pil, make_image, make_bounding_boxes, make_segmentation_mask]
+    )
+    @pytest.mark.parametrize(
+        "make_input2", [make_image_tensor, make_image_pil, make_image, make_bounding_boxes, make_segmentation_mask]
+    )
+    @pytest.mark.parametrize("query", [transforms.query_size, transforms.query_chw])
+    def test_different_sizes(self, make_input1, make_input2, query):
+        input1 = make_input1((10, 10))
+        input2 = make_input2((20, 20))
+        if query is transforms.query_chw and not all(
+            transforms.check_type(inpt, (is_pure_tensor, tv_tensors.Image, PIL.Image.Image, tv_tensors.Video))
+            for inpt in (input1, input2)
+        ):
+            return
+        with pytest.raises(ValueError, match="Found multiple"):
+            query([input1, input2])
+
+    @pytest.mark.parametrize("query", [transforms.query_size, transforms.query_chw])
+    def test_no_valid_input(self, query):
+        with pytest.raises(TypeError, match="No image"):
+            query(["blah"])

test/test_transforms_v2.py

@@ -44,10 +44,12 @@ torch::Tensor decode_webp(
 
   auto decoded_data =
       decoding_func(encoded_data_p, encoded_data_size, &width, &height);
+
   TORCH_CHECK(decoded_data != nullptr, "WebPDecodeRGB[A] failed.");
 
+  auto deleter = [decoded_data](void*) { WebPFree(decoded_data); };
   auto out = torch::from_blob(
-      decoded_data, {height, width, num_channels}, torch::kUInt8);
+      decoded_data, {height, width, num_channels}, deleter, torch::kUInt8);
 
   return out.permute({2, 0, 1});
 }

torchvision/csrc/io/image/cpu/decode_webp.cpp

@@ -72,6 +72,7 @@
     "QMNIST",
     "MNIST",
     "KMNIST",
+    "MovingMNIST",
     "StanfordCars",
     "STL10",
     "SUN397",

torchvision/datasets/__init__.py

@@ -588,7 +588,6 @@ def _download_dataset(self, root: Union[str, Path]) -> None:
                 for calibration in ["perfect", "imperfect"]:
                     scene_name = f"{split_scene}-{calibration}"
                     scene_url = f"{base_url}/{scene_name}.zip"
-                    print(f"Downloading {scene_url}")
                     # download the scene only if it doesn't exist
                     if not (split_root / scene_name).exists():
                         download_and_extract_archive(

torchvision/datasets/_stereo_matching.py

@@ -130,7 +130,6 @@ def __len__(self) -> int:
 
     def download(self) -> None:
         if self._check_integrity():
-            print("Files already downloaded and verified")
             return
 
         download_and_extract_archive(
@@ -231,7 +230,6 @@ def __len__(self) -> int:
 
     def download(self) -> None:
         if self._check_integrity():
-            print("Files already downloaded and verified")
             return
 
         download_and_extract_archive(

torchvision/datasets/caltech.py

@@ -105,7 +105,7 @@ def __init__(
         if mask == slice(None):  # if split == "all"
             self.filename = splits.index
         else:
-            self.filename = [splits.index[i] for i in torch.squeeze(torch.nonzero(mask))]
+            self.filename = [splits.index[i] for i in torch.squeeze(torch.nonzero(mask))]  # type: ignore[arg-type]
         self.identity = identity.data[mask]
         self.bbox = bbox.data[mask]
         self.landmarks_align = landmarks_align.data[mask]
@@ -148,7 +148,6 @@ def _check_integrity(self) -> bool:
 
     def download(self) -> None:
         if self._check_integrity():
-            print("Files already downloaded and verified")
             return
 
         for (file_id, md5, filename) in self.file_list: