[fbsync] remove datapoints compatibility for prototype datasets (#7154)

Reviewed By: vmoens

Differential Revision: D44416260

fbshipit-source-id: 9437edffc8a7ccf08f381c5147d9a8f3e18530a3

Author: NicolasHug

test/test_prototype_datasets_builtin.py

@@ -21,6 +21,7 @@
 from torchdata.datapipes.utils import StreamWrapper
 from torchvision._utils import sequence_to_str
 from torchvision.prototype import datapoints, datasets, transforms
+from torchvision.prototype.datasets.utils import EncodedImage
 from torchvision.prototype.datasets.utils._internal import INFINITE_BUFFER_SIZE
 
 
@@ -136,18 +137,21 @@ def make_msg_and_close(head):
             raise AssertionError(make_msg_and_close("The following streams were not closed after a full iteration:"))
 
     @parametrize_dataset_mocks(DATASET_MOCKS)
-    def test_no_simple_tensors(self, dataset_mock, config):
+    def test_no_unaccompanied_simple_tensors(self, dataset_mock, config):
         dataset, _ = dataset_mock.load(config)
+        sample = next_consume(iter(dataset))
 
         simple_tensors = {
-            key
-            for key, value in next_consume(iter(dataset)).items()
-            if torchvision.prototype.transforms.utils.is_simple_tensor(value)
+            key for key, value in sample.items() if torchvision.prototype.transforms.utils.is_simple_tensor(value)
         }
-        if simple_tensors:
+
+        if simple_tensors and not any(
+            isinstance(item, (datapoints.Image, datapoints.Video, EncodedImage)) for item in sample.values()
+        ):
             raise AssertionError(
                 f"The values of key(s) "
-                f"{sequence_to_str(sorted(simple_tensors), separate_last='and ')} contained simple tensors."
+                f"{sequence_to_str(sorted(simple_tensors), separate_last='and ')} contained simple tensors, "
+                f"but didn't find any (encoded) image or video."
             )
 
     @parametrize_dataset_mocks(DATASET_MOCKS)

torchvision/prototype/datapoints/_datapoint.py

@@ -29,26 +29,9 @@ def _to_tensor(
             requires_grad = data.requires_grad if isinstance(data, torch.Tensor) else False
         return torch.as_tensor(data, dtype=dtype, device=device).requires_grad_(requires_grad)
 
-    # FIXME: this is just here for BC with the prototype datasets. Some datasets use the Datapoint directly to have a
-    #  a no-op input for the prototype transforms. For this use case, we can't use plain tensors, since they will be
-    #  interpreted as images. We should decide if we want a public no-op datapoint like `GenericDatapoint` or make this
-    #  one public again.
-    def __new__(
-        cls,
-        data: Any,
-        dtype: Optional[torch.dtype] = None,
-        device: Optional[Union[torch.device, str, int]] = None,
-        requires_grad: Optional[bool] = None,
-    ) -> Datapoint:
-        tensor = cls._to_tensor(data, dtype=dtype, device=device, requires_grad=requires_grad)
-        return tensor.as_subclass(Datapoint)
-
     @classmethod
     def wrap_like(cls: Type[D], other: D, tensor: torch.Tensor) -> D:
-        # FIXME: this is just here for BC with the prototype datasets. See __new__ for details. If that is resolved,
-        #  this method should be made abstract
-        # raise NotImplementedError
-        return tensor.as_subclass(cls)
+        raise NotImplementedError
 
     _NO_WRAPPING_EXCEPTIONS = {
         torch.Tensor.clone: lambda cls, input, output: cls.wrap_like(input, output),

torchvision/prototype/datasets/_builtin/caltech.py

@@ -3,9 +3,10 @@
 from typing import Any, BinaryIO, Dict, List, Tuple, Union
 
 import numpy as np
+
+import torch
 from torchdata.datapipes.iter import Filter, IterDataPipe, IterKeyZipper, Mapper
 from torchvision.prototype.datapoints import BoundingBox, Label
-from torchvision.prototype.datapoints._datapoint import Datapoint
 from torchvision.prototype.datasets.utils import Dataset, EncodedImage, GDriveResource, OnlineResource
 from torchvision.prototype.datasets.utils._internal import (
     hint_sharding,
@@ -115,7 +116,7 @@ def _prepare_sample(
                 format="xyxy",
                 spatial_size=image.spatial_size,
             ),
-            contour=Datapoint(ann["obj_contour"].T),
+            contour=torch.as_tensor(ann["obj_contour"].T),
         )
 
     def _datapipe(self, resource_dps: List[IterDataPipe]) -> IterDataPipe[Dict[str, Any]]:

torchvision/prototype/datasets/_builtin/celeba.py

@@ -2,9 +2,9 @@
 import pathlib
 from typing import Any, BinaryIO, Dict, Iterator, List, Optional, Sequence, Tuple, Union
 
+import torch
 from torchdata.datapipes.iter import Filter, IterDataPipe, IterKeyZipper, Mapper, Zipper
 from torchvision.prototype.datapoints import BoundingBox, Label
-from torchvision.prototype.datapoints._datapoint import Datapoint
 from torchvision.prototype.datasets.utils import Dataset, EncodedImage, GDriveResource, OnlineResource
 from torchvision.prototype.datasets.utils._internal import (
     getitem,
@@ -149,7 +149,7 @@ def _prepare_sample(
                 spatial_size=image.spatial_size,
             ),
             landmarks={
-                landmark: Datapoint((int(landmarks[f"{landmark}_x"]), int(landmarks[f"{landmark}_y"])))
+                landmark: torch.tensor((int(landmarks[f"{landmark}_x"]), int(landmarks[f"{landmark}_y"])))
                 for landmark in {key[:-2] for key in landmarks.keys()}
             },
         )

torchvision/prototype/datasets/_builtin/coco.py

@@ -15,7 +15,6 @@
     UnBatcher,
 )
 from torchvision.prototype.datapoints import BoundingBox, Label, Mask
-from torchvision.prototype.datapoints._datapoint import Datapoint
 from torchvision.prototype.datasets.utils import Dataset, EncodedImage, HttpResource, OnlineResource
 from torchvision.prototype.datasets.utils._internal import (
     getitem,
@@ -124,8 +123,8 @@ def _decode_instances_anns(self, anns: List[Dict[str, Any]], image_meta: Dict[st
                     ]
                 )
             ),
-            areas=Datapoint([ann["area"] for ann in anns]),
-            crowds=Datapoint([ann["iscrowd"] for ann in anns], dtype=torch.bool),
+            areas=torch.as_tensor([ann["area"] for ann in anns]),
+            crowds=torch.as_tensor([ann["iscrowd"] for ann in anns], dtype=torch.bool),
             bounding_boxes=BoundingBox(
                 [ann["bbox"] for ann in anns],
                 format="xywh",

torchvision/prototype/datasets/_builtin/cub200.py

@@ -3,6 +3,7 @@
 import pathlib
 from typing import Any, BinaryIO, Callable, Dict, List, Optional, Tuple, Union
 
+import torch
 from torchdata.datapipes.iter import (
     CSVDictParser,
     CSVParser,
@@ -15,7 +16,6 @@
 )
 from torchdata.datapipes.map import IterToMapConverter
 from torchvision.prototype.datapoints import BoundingBox, Label
-from torchvision.prototype.datapoints._datapoint import Datapoint
 from torchvision.prototype.datasets.utils import Dataset, EncodedImage, GDriveResource, OnlineResource
 from torchvision.prototype.datasets.utils._internal import (
     getitem,
@@ -162,7 +162,7 @@ def _2010_prepare_ann(
                 format="xyxy",
                 spatial_size=spatial_size,
             ),
-            segmentation=Datapoint(content["seg"]),
+            segmentation=torch.as_tensor(content["seg"]),
         )
 
     def _prepare_sample(

torchvision/prototype/datasets/_builtin/sbd.py

@@ -3,8 +3,8 @@
 from typing import Any, BinaryIO, cast, Dict, List, Optional, Tuple, Union
 
 import numpy as np
+import torch
 from torchdata.datapipes.iter import Demultiplexer, Filter, IterDataPipe, IterKeyZipper, LineReader, Mapper
-from torchvision.prototype.datapoints._datapoint import Datapoint
 from torchvision.prototype.datasets.utils import Dataset, EncodedImage, HttpResource, OnlineResource
 from torchvision.prototype.datasets.utils._internal import (
     getitem,
@@ -92,8 +92,10 @@ def _prepare_sample(self, data: Tuple[Tuple[Any, Tuple[str, BinaryIO]], Tuple[st
             image=EncodedImage.from_file(image_buffer),
             ann_path=ann_path,
             # the boundaries are stored in sparse CSC format, which is not supported by PyTorch
-            boundaries=Datapoint(np.stack([raw_boundary.toarray() for raw_boundary in anns["Boundaries"].item()])),
-            segmentation=Datapoint(anns["Segmentation"].item()),
+            boundaries=torch.as_tensor(
+                np.stack([raw_boundary.toarray() for raw_boundary in anns["Boundaries"].item()])
+            ),
+            segmentation=torch.as_tensor(anns["Segmentation"].item()),
         )
 
     def _datapipe(self, resource_dps: List[IterDataPipe]) -> IterDataPipe[Dict[str, Any]]: