Merge branch 'main' into tests/flaky_stochastic_depth

Author: datumbox

torchvision/prototype/datasets/_builtin/caltech.py

@@ -21,16 +21,14 @@
     OnlineResource,
     DatasetType,
 )
-from torchvision.prototype.datasets.utils._internal import INFINITE_BUFFER_SIZE, BUILTIN_DIR, read_mat
+from torchvision.prototype.datasets.utils._internal import INFINITE_BUFFER_SIZE, read_mat
 
 
 class Caltech101(Dataset):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "caltech101",
             type=DatasetType.IMAGE,
-            categories=BUILTIN_DIR / "caltech101.categories",
             homepage="http://www.vision.caltech.edu/Image_Datasets/Caltech101",
         )
 
@@ -144,12 +142,10 @@ def _generate_categories(self, root: pathlib.Path) -> List[str]:
 
 
 class Caltech256(Dataset):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "caltech256",
             type=DatasetType.IMAGE,
-            categories=BUILTIN_DIR / "caltech256.categories",
             homepage="http://www.vision.caltech.edu/Image_Datasets/Caltech256",
         )
 

torchvision/prototype/datasets/_builtin/celeba.py

@@ -58,8 +58,7 @@ def __iter__(self) -> Iterator[Tuple[str, Dict[str, str]]]:
 
 
 class CelebA(Dataset):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "celeba",
             type=DatasetType.IMAGE,

torchvision/prototype/datasets/_builtin/cifar.py

@@ -25,7 +25,6 @@
 )
 from torchvision.prototype.datasets.utils._internal import (
     INFINITE_BUFFER_SIZE,
-    BUILTIN_DIR,
     image_buffer_from_array,
     path_comparator,
 )
@@ -110,12 +109,10 @@ def _is_data_file(self, data: Tuple[str, Any], *, config: DatasetConfig) -> bool
         path = pathlib.Path(data[0])
         return path.name.startswith("data" if config.split == "train" else "test")
 
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "cifar10",
             type=DatasetType.RAW,
-            categories=BUILTIN_DIR / "cifar10.categories",
             homepage="https://www.cs.toronto.edu/~kriz/cifar.html",
         )
 
@@ -137,12 +134,10 @@ def _is_data_file(self, data: Tuple[str, Any], *, config: DatasetConfig) -> bool
         path = pathlib.Path(data[0])
         return path.name == cast(str, config.split)
 
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "cifar100",
             type=DatasetType.RAW,
-            categories=BUILTIN_DIR / "cifar100.categories",
             homepage="https://www.cs.toronto.edu/~kriz/cifar.html",
             valid_options=dict(
                 split=("train", "test"),

torchvision/prototype/datasets/_builtin/coco.py

@@ -35,8 +35,7 @@
 
 
 class Coco(Dataset):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "coco",
             type=DatasetType.IMAGE,

torchvision/prototype/datasets/_builtin/imagenet.py

@@ -25,8 +25,7 @@
 
 
 class ImageNet(Dataset):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         name = "imagenet"
         categories, wnids = zip(*DatasetInfo.read_categories_file(BUILTIN_DIR / f"{name}.categories"))
 

torchvision/prototype/datasets/_builtin/mnist.py

@@ -151,8 +151,7 @@ def _make_datapipe(
 
 
 class MNIST(_MNISTBase):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "mnist",
             type=DatasetType.RAW,
@@ -182,8 +181,7 @@ def _files_and_checksums(self, config: DatasetConfig) -> Tuple[Tuple[str, str],
 
 
 class FashionMNIST(MNIST):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "fashionmnist",
             type=DatasetType.RAW,
@@ -215,8 +213,7 @@ def info(self) -> DatasetInfo:
 
 
 class KMNIST(MNIST):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "kmnist",
             type=DatasetType.RAW,
@@ -237,8 +234,7 @@ def info(self) -> DatasetInfo:
 
 
 class EMNIST(_MNISTBase):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "emnist",
             type=DatasetType.RAW,
@@ -335,8 +331,7 @@ def _make_datapipe(
 
 
 class QMNIST(_MNISTBase):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "qmnist",
             type=DatasetType.RAW,

torchvision/prototype/datasets/_builtin/sbd.py

@@ -25,7 +25,6 @@
 )
 from torchvision.prototype.datasets.utils._internal import (
     INFINITE_BUFFER_SIZE,
-    BUILTIN_DIR,
     read_mat,
     getitem,
     path_accessor,
@@ -34,12 +33,10 @@
 
 
 class SBD(Dataset):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "sbd",
             type=DatasetType.IMAGE,
-            categories=BUILTIN_DIR / "caltech256.categories",
             homepage="http://home.bharathh.info/pubs/codes/SBD/download.html",
             valid_options=dict(
                 split=("train", "val", "train_noval"),

torchvision/prototype/datasets/_builtin/voc.py

@@ -35,8 +35,7 @@
 
 
 class VOC(Dataset):
-    @property
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         return DatasetInfo(
             "voc",
             type=DatasetType.IMAGE,

torchvision/prototype/datasets/utils/_dataset.py

@@ -12,7 +12,7 @@
     sequence_to_str,
 )
 
-from ._internal import FrozenBunch, make_repr
+from ._internal import FrozenBunch, make_repr, BUILTIN_DIR
 from ._resource import OnlineResource
 
 
@@ -42,8 +42,9 @@ def __init__(
         self.type = DatasetType[type.upper()] if isinstance(type, str) else type
 
         if categories is None:
-            categories = []
-        elif isinstance(categories, int):
+            path = BUILTIN_DIR / f"{self.name}.categories"
+            categories = path if path.exists() else []
+        if isinstance(categories, int):
             categories = [str(label) for label in range(categories)]
         elif isinstance(categories, (str, pathlib.Path)):
             path = pathlib.Path(categories).expanduser().resolve()
@@ -112,11 +113,17 @@ def __repr__(self) -> str:
 
 
 class Dataset(abc.ABC):
-    @property
+    def __init__(self) -> None:
+        self._info = self._make_info()
+
     @abc.abstractmethod
-    def info(self) -> DatasetInfo:
+    def _make_info(self) -> DatasetInfo:
         pass
 
+    @property
+    def info(self) -> DatasetInfo:
+        return self._info
+
     @property
     def name(self) -> str:
         return self.info.name