Compatibility layer between stable datasets and prototype transforms

test/datasets_utils.py

@@ -581,6 +581,28 @@ def test_transforms(self, config):
 
                 mock.assert_called()
 
+    @test_all_configs
+    def test_transforms_v2_wrapper(self, config):
+        # This is stable test, so we can't depend on prototype stuff
+        try:
+            from torchvision.prototype.datapoints import wrap_dataset_for_transforms_v2
+        except ImportError:
+            return
+
+        try:
+            with self.create_dataset(config) as (dataset, _):
+                wrapped_dataset = wrap_dataset_for_transforms_v2(dataset)
+                wrapped_sample = wrapped_dataset[0]
+                assert wrapped_sample is not None
+        except ValueError as error:
+            if str(error).startswith(f"No wrapper exist for dataset class {type(dataset).__name__}"):
+                return
+            raise error
+        except RuntimeError as error:
+            if "currently not supported by this wrapper" in str(error):
+                return
+            raise error
+
 
 class ImageDatasetTestCase(DatasetTestCase):
     """Abstract base class for image dataset testcases.
@@ -662,6 +684,15 @@ def wrapper(tmpdir, config):
 
         return wrapper
 
+    @test_all_configs
+    def test_transforms_v2_wrapper(self, config):
+        # `output_format == "THWC"` is not supported by the wrapper. Thus, we skip the `config` if it is set explicitly
+        # or use the supported `"TCHW"`
+        if config.setdefault("output_format", "TCHW") == "THWC":
+            return
+
+        super().test_transforms_v2_wrapper.__wrapped__(self, config)
+
 
 def create_image_or_video_tensor(size: Sequence[int]) -> torch.Tensor:
     r"""Create a random uint8 tensor.

test/test_datasets.py

@@ -763,11 +763,19 @@ def _create_annotation_file(self, root, name, file_names, num_annotations_per_im
         return info
 
     def _create_annotations(self, image_ids, num_annotations_per_image):
-        annotations = datasets_utils.combinations_grid(
-            image_id=image_ids, bbox=([1.0, 2.0, 3.0, 4.0],) * num_annotations_per_image
-        )
-        for id, annotation in enumerate(annotations):
-            annotation["id"] = id
+        annotations = []
+        annotion_id = 0
+        for image_id in itertools.islice(itertools.cycle(image_ids), len(image_ids) * num_annotations_per_image):
+            annotations.append(
+                dict(
+                    image_id=image_id,
+                    id=annotion_id,
+                    bbox=torch.rand(4).tolist(),
+                    segmentation=[torch.rand(8).tolist()],
+                    category_id=int(torch.randint(91, ())),
+                )
+            )
+            annotion_id += 1
         return annotations, dict()
 
     def _create_json(self, root, name, content):

torchvision/prototype/datapoints/__init__.py

@@ -4,3 +4,5 @@
 from ._label import Label, OneHotLabel
 from ._mask import Mask
 from ._video import TensorVideoType, TensorVideoTypeJIT, Video, VideoType, VideoTypeJIT
+
+from ._dataset_wrapper import wrap_dataset_for_transforms_v2  # type: ignore[attr-defined]  # usort: skip