From d47ede984b08b5f6bf7d5b867a92b78f9c6c6cfd Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Thu, 30 Sep 2021 15:54:48 +0200
Subject: [PATCH 01/11] make mypy more strict for prototype datasets

---
 mypy.ini                                         | 16 ++++++++++++++++
 .../prototype/datasets/_builtin/caltech.py       | 13 ++++++++-----
 torchvision/prototype/datasets/_folder.py        |  6 +++---
 torchvision/prototype/datasets/decoder.py        |  3 ++-
 torchvision/prototype/datasets/utils/_dataset.py | 12 +++++++-----
 .../prototype/datasets/utils/_resource.py        |  2 +-
 6 files changed, 37 insertions(+), 15 deletions(-)

diff --git a/mypy.ini b/mypy.ini
index dac60e11ce0..2016b93fd2d 100644
--- a/mypy.ini
+++ b/mypy.ini
@@ -4,6 +4,22 @@ files = torchvision
 show_error_codes = True
 pretty = True
 
+[mypy-torchvision.prototype.*]
+
+; untyped definitions and calls
+disallow_untyped_defs = True
+
+; None and Optional handling
+no_implicit_optional = True
+
+; warnings
+warn_unused_ignores = True
+warn_return_any = True
+warn_unreachable = True
+
+; miscellaneous strictness flags
+allow_redefinition = True
+
 [mypy-torchvision.io._video_opt.*]
 
 ignore_errors = True
diff --git a/torchvision/prototype/datasets/_builtin/caltech.py b/torchvision/prototype/datasets/_builtin/caltech.py
index 7f6021522c8..8a608f54bd1 100644
--- a/torchvision/prototype/datasets/_builtin/caltech.py
+++ b/torchvision/prototype/datasets/_builtin/caltech.py
@@ -84,7 +84,10 @@ def _anns_key_fn(self, data: Tuple[str, Any]) -> Tuple[str, str]:
         return category, id
 
     def _collate_and_decode_sample(
-        self, data, *, decoder: Optional[Callable[[io.IOBase], torch.Tensor]]
+        self,
+        data: Tuple[Tuple[str, str], Tuple[str, io.IOBase], Tuple[str, io.IOBase]],
+        *,
+        decoder: Optional[Callable[[io.IOBase], torch.Tensor]],
     ) -> Dict[str, Any]:
         key, image_data, ann_data = data
         category, _ = key
@@ -119,12 +122,12 @@ def _make_datapipe(
         images_dp, anns_dp = resource_dps
 
         images_dp = TarArchiveReader(images_dp)
-        images_dp = Filter(images_dp, self._is_not_background_image)
+        images_dp: IterDataPipe = Filter(images_dp, self._is_not_background_image)
         # FIXME: add this after https://github.com/pytorch/pytorch/issues/65808 is resolved
         # images_dp = Shuffler(images_dp, buffer_size=INFINITE_BUFFER_SIZE)
 
         anns_dp = TarArchiveReader(anns_dp)
-        anns_dp = Filter(anns_dp, self._is_ann)
+        anns_dp: IterDataPipe = Filter(anns_dp, self._is_ann)
 
         dp = KeyZipper(
             images_dp,
@@ -139,7 +142,7 @@ def _make_datapipe(
     def generate_categories_file(self, root: Union[str, pathlib.Path]) -> None:
         dp = self.resources(self.default_config)[0].to_datapipe(pathlib.Path(root) / self.name)
         dp = TarArchiveReader(dp)
-        dp = Filter(dp, self._is_not_background_image)
+        dp: IterDataPipe = Filter(dp, self._is_not_background_image)
         dir_names = {pathlib.Path(path).parent.name for path, _ in dp}
         create_categories_file(HERE, self.name, sorted(dir_names))
 
@@ -188,7 +191,7 @@ def _make_datapipe(
     ) -> IterDataPipe[Dict[str, Any]]:
         dp = resource_dps[0]
         dp = TarArchiveReader(dp)
-        dp = Filter(dp, self._is_not_rogue_file)
+        dp: IterDataPipe = Filter(dp, self._is_not_rogue_file)
         # FIXME: add this after https://github.com/pytorch/pytorch/issues/65808 is resolved
         # dp = Shuffler(dp, buffer_size=INFINITE_BUFFER_SIZE)
         return Mapper(dp, self._collate_and_decode_sample, fn_kwargs=dict(decoder=decoder))
diff --git a/torchvision/prototype/datasets/_folder.py b/torchvision/prototype/datasets/_folder.py
index 5626f68650f..0e8bb56fc36 100644
--- a/torchvision/prototype/datasets/_folder.py
+++ b/torchvision/prototype/datasets/_folder.py
@@ -26,7 +26,7 @@ def _collate_and_decode_data(
     *,
     root: pathlib.Path,
     categories: List[str],
-    decoder,
+    decoder: Optional[Callable[[io.IOBase], torch.Tensor]],
 ) -> Dict[str, Any]:
     path, buffer = data
     data = decoder(buffer) if decoder else buffer
@@ -50,8 +50,8 @@ def from_data_folder(
     root = pathlib.Path(root).expanduser().resolve()
     categories = sorted(entry.name for entry in os.scandir(root) if entry.is_dir())
     masks: Union[List[str], str] = [f"*.{ext}" for ext in valid_extensions] if valid_extensions is not None else ""
-    dp: IterDataPipe = FileLister(str(root), recursive=recursive, masks=masks)
-    dp = Filter(dp, _is_not_top_level_file, fn_kwargs=dict(root=root))
+    dp = FileLister(str(root), recursive=recursive, masks=masks)
+    dp: IterDataPipe = Filter(dp, _is_not_top_level_file, fn_kwargs=dict(root=root))
     dp = Shuffler(dp, buffer_size=INFINITE_BUFFER_SIZE)
     dp = FileLoader(dp)
     return (
diff --git a/torchvision/prototype/datasets/decoder.py b/torchvision/prototype/datasets/decoder.py
index d4897bebc91..f280d81a36f 100644
--- a/torchvision/prototype/datasets/decoder.py
+++ b/torchvision/prototype/datasets/decoder.py
@@ -1,4 +1,5 @@
 import io
+from typing import cast
 
 import PIL.Image
 import torch
@@ -9,4 +10,4 @@
 
 
 def pil(buffer: io.IOBase, mode: str = "RGB") -> torch.Tensor:
-    return pil_to_tensor(PIL.Image.open(buffer).convert(mode.upper()))
+    return cast(torch.Tensor, pil_to_tensor(PIL.Image.open(buffer).convert(mode.upper())))
diff --git a/torchvision/prototype/datasets/utils/_dataset.py b/torchvision/prototype/datasets/utils/_dataset.py
index 19fb3b1d596..1937f51f9d5 100644
--- a/torchvision/prototype/datasets/utils/_dataset.py
+++ b/torchvision/prototype/datasets/utils/_dataset.py
@@ -15,6 +15,8 @@
     NoReturn,
     Iterable,
     Tuple,
+    Iterator,
+    cast,
 )
 
 import torch
@@ -27,7 +29,7 @@
 from ._resource import OnlineResource
 
 
-def make_repr(name: str, items: Iterable[Tuple[str, Any]]):
+def make_repr(name: str, items: Iterable[Tuple[str, Any]]) -> str:
     def to_str(sep: str) -> str:
         return sep.join([f"{key}={value}" for key, value in items])
 
@@ -46,7 +48,7 @@ def to_str(sep: str) -> str:
 
 
 class DatasetConfig(Mapping):
-    def __init__(self, *args, **kwargs):
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
         data = dict(*args, **kwargs)
         self.__dict__["__data__"] = data
         self.__dict__["__final_hash__"] = hash(tuple(data.items()))
@@ -54,10 +56,10 @@ def __init__(self, *args, **kwargs):
     def __getitem__(self, name: str) -> Any:
         return self.__dict__["__data__"][name]
 
-    def __iter__(self):
+    def __iter__(self) -> Iterator[Any]:
         return iter(self.__dict__["__data__"].keys())
 
-    def __len__(self):
+    def __len__(self) -> int:
         return len(self.__dict__["__data__"])
 
     def __getattr__(self, name: str) -> Any:
@@ -81,7 +83,7 @@ def __delattr__(self, item: Any) -> NoReturn:
         raise RuntimeError(f"'{type(self).__name__}' object is immutable")
 
     def __hash__(self) -> int:
-        return self.__dict__["__final_hash__"]
+        return cast(int, self.__dict__["__final_hash__"])
 
     def __eq__(self, other: Any) -> bool:
         if not isinstance(other, DatasetConfig):
diff --git a/torchvision/prototype/datasets/utils/_resource.py b/torchvision/prototype/datasets/utils/_resource.py
index 3f372d0f5b7..522fd4e3c46 100644
--- a/torchvision/prototype/datasets/utils/_resource.py
+++ b/torchvision/prototype/datasets/utils/_resource.py
@@ -8,7 +8,7 @@
 
 
 # FIXME
-def compute_sha256(_) -> str:
+def compute_sha256(path: pathlib.Path) -> str:
     return ""
 
 

From be3babf3ea6be13fd0c0ff6c17038871c0635d7e Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Tue, 5 Oct 2021 08:29:46 +0200
Subject: [PATCH 02/11] fix code format

---
 torchvision/prototype/datasets/_builtin/caltech.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/torchvision/prototype/datasets/_builtin/caltech.py b/torchvision/prototype/datasets/_builtin/caltech.py
index 50b0b96a88f..7c25037691f 100644
--- a/torchvision/prototype/datasets/_builtin/caltech.py
+++ b/torchvision/prototype/datasets/_builtin/caltech.py
@@ -121,8 +121,7 @@ def _make_datapipe(
 
         images_dp = TarArchiveReader(images_dp)
         images_dp: IterDataPipe = Filter(images_dp, self._is_not_background_image)
-        # FIXME: add this after https://github.com/pytorch/pytorch/issues/65808 is resolved
-        # images_dp = Shuffler(images_dp, buffer_size=INFINITE_BUFFER_SIZE)
+        images_dp = Shuffler(images_dp, buffer_size=INFINITE_BUFFER_SIZE)
 
         anns_dp = TarArchiveReader(anns_dp)
         anns_dp: IterDataPipe = Filter(anns_dp, self._is_ann)
@@ -190,8 +189,7 @@ def _make_datapipe(
         dp = resource_dps[0]
         dp = TarArchiveReader(dp)
         dp: IterDataPipe = Filter(dp, self._is_not_rogue_file)
-        # FIXME: add this after https://github.com/pytorch/pytorch/issues/65808 is resolved
-        # dp = Shuffler(dp, buffer_size=INFINITE_BUFFER_SIZE)
+        dp = Shuffler(dp, buffer_size=INFINITE_BUFFER_SIZE)
         return Mapper(dp, self._collate_and_decode_sample, fn_kwargs=dict(decoder=decoder))
 
     def generate_categories_file(self, root: Union[str, pathlib.Path]) -> None:

From b437fa31b9a3527ba6c76a6b6d09e95263801caf Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Wed, 20 Oct 2021 15:54:47 +0200
Subject: [PATCH 03/11] apply strictness only to datasets

---
 mypy.ini | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mypy.ini b/mypy.ini
index ffb113081f6..6069eece9bc 100644
--- a/mypy.ini
+++ b/mypy.ini
@@ -5,7 +5,7 @@ show_error_codes = True
 pretty = True
 allow_redefinition = True
 
-[mypy-torchvision.prototype.*]
+[mypy-torchvision.prototype.datasets.*]
 
 ; untyped definitions and calls
 disallow_untyped_defs = True

From 182a4ea6ffac6ac9e78d12215e67e056937ad497 Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Wed, 20 Oct 2021 15:54:58 +0200
Subject: [PATCH 04/11] fix more mypy issues

---
 .../prototype/datasets/_builtin/celeba.py     | 28 ++++++++++---------
 .../prototype/datasets/_builtin/cifar.py      | 10 +++----
 .../prototype/datasets/_builtin/mnist.py      | 20 ++++++-------
 .../prototype/datasets/_builtin/sbd.py        | 22 +++++++++------
 .../prototype/datasets/_builtin/voc.py        | 10 +++++--
 torchvision/prototype/datasets/benchmark.py   |  2 ++
 .../datasets/generate_category_files.py       |  4 ++-
 .../prototype/datasets/utils/_internal.py     |  6 ++--
 8 files changed, 59 insertions(+), 43 deletions(-)

diff --git a/torchvision/prototype/datasets/_builtin/celeba.py b/torchvision/prototype/datasets/_builtin/celeba.py
index d86eaf27fab..1a604cd756e 100644
--- a/torchvision/prototype/datasets/_builtin/celeba.py
+++ b/torchvision/prototype/datasets/_builtin/celeba.py
@@ -1,6 +1,6 @@
 import csv
 import io
-from typing import Any, Callable, Dict, List, Optional, Tuple, Mapping, Union
+from typing import Any, Callable, Dict, List, Optional, Tuple, Mapping, Union, Iterator
 
 import torch
 from torchdata.datapipes.iter import (
@@ -23,18 +23,20 @@
 from torchvision.prototype.datasets.utils._internal import INFINITE_BUFFER_SIZE, getitem, path_accessor
 
 
+csv.register_dialect("celeba", delimiter=" ", skipinitialspace=True)
+
+
 class CelebACSVParser(IterDataPipe):
     def __init__(
         self,
-        datapipe,
+        datapipe: IterDataPipe[Tuple[str, io.IOBase]],
         *,
-        has_header,
-    ):
+        has_header: bool,
+    ) -> None:
         self.datapipe = datapipe
         self.has_header = has_header
-        self._fmtparams = dict(delimiter=" ", skipinitialspace=True)
 
-    def __iter__(self):
+    def __iter__(self) -> Iterator[Tuple[str, Union[Dict[str, str], List[str]]]]:
         for _, file in self.datapipe:
             file = (line.decode() for line in file)
 
@@ -42,18 +44,18 @@ def __iter__(self):
                 # The first row is skipped, because it only contains the number of samples
                 next(file)
 
-                # Empty field names are filtered out, because some files have an extr white space after the header
+                # Empty field names are filtered out, because some files have an extra white space after the header
                 # line, which is recognized as extra column
-                fieldnames = [name for name in next(csv.reader([next(file)], **self._fmtparams)) if name]
+                fieldnames = [name for name in next(csv.reader([next(file)], dialect="celeba")) if name]
                 # Some files do not include a label for the image ID column
                 if fieldnames[0] != "image_id":
                     fieldnames.insert(0, "image_id")
 
-                for line in csv.DictReader(file, fieldnames=fieldnames, **self._fmtparams):
-                    yield line.pop("image_id"), line
+                for line_dict in csv.DictReader(file, fieldnames=fieldnames, dialect="celeba"):
+                    yield line_dict.pop("image_id"), line_dict
             else:
-                for line in csv.reader(file, **self._fmtparams):
-                    yield line[0], line[1:]
+                for line_list in csv.reader(file, dialect="celeba"):
+                    yield line_list[0], line_list[1:]
 
 
 class CelebA(Dataset):
@@ -104,7 +106,7 @@ def resources(self, config: DatasetConfig) -> List[OnlineResource]:
         "2": "test",
     }
 
-    def _filter_split(self, data: Tuple[str, str], *, split):
+    def _filter_split(self, data: Tuple[str, str], *, split: str) -> bool:
         _, split_id = data
         return self._SPLIT_ID_TO_NAME[split_id[0]] == split
 
diff --git a/torchvision/prototype/datasets/_builtin/cifar.py b/torchvision/prototype/datasets/_builtin/cifar.py
index 4fbd993d311..4949c56f6cb 100644
--- a/torchvision/prototype/datasets/_builtin/cifar.py
+++ b/torchvision/prototype/datasets/_builtin/cifar.py
@@ -3,7 +3,7 @@
 import io
 import pathlib
 import pickle
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union, Iterator
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union, Iterator, cast
 
 import numpy as np
 import torch
@@ -56,7 +56,7 @@ def _is_data_file(self, data: Tuple[str, io.IOBase], *, config: DatasetConfig) -
 
     def _unpickle(self, data: Tuple[str, io.BytesIO]) -> Dict[str, Any]:
         _, file = data
-        return pickle.load(file, encoding="latin1")
+        return cast(Dict[str, Any], pickle.load(file, encoding="latin1"))
 
     def _collate_and_decode(
         self,
@@ -98,7 +98,7 @@ def _generate_categories(self, root: pathlib.Path) -> List[str]:
         dp = TarArchiveReader(dp)
         dp: IterDataPipe = Filter(dp, path_comparator("name", self._META_FILE_NAME))
         dp: IterDataPipe = Mapper(dp, self._unpickle)
-        return next(iter(dp))[self._CATEGORIES_KEY]
+        return cast(List[str], next(iter(dp))[self._CATEGORIES_KEY])
 
 
 class Cifar10(_CifarBase):
@@ -133,9 +133,9 @@ class Cifar100(_CifarBase):
     _META_FILE_NAME = "meta"
     _CATEGORIES_KEY = "fine_label_names"
 
-    def _is_data_file(self, data: Tuple[str, io.IOBase], *, config: DatasetConfig) -> bool:
+    def _is_data_file(self, data: Tuple[str, Any], *, config: DatasetConfig) -> bool:
         path = pathlib.Path(data[0])
-        return path.name == config.split
+        return path.name == cast(str, config.split)
 
     @property
     def info(self) -> DatasetInfo:
diff --git a/torchvision/prototype/datasets/_builtin/mnist.py b/torchvision/prototype/datasets/_builtin/mnist.py
index 2413a2fb084..9778c920f37 100644
--- a/torchvision/prototype/datasets/_builtin/mnist.py
+++ b/torchvision/prototype/datasets/_builtin/mnist.py
@@ -54,7 +54,7 @@ def __init__(self, datapipe: IterDataPipe, *, start: Optional[int], stop: Option
         self.stop = stop
 
     @staticmethod
-    def _decode(bytes):
+    def _decode(bytes: bytes) -> int:
         return int(codecs.encode(bytes, "hex"), 16)
 
     def __iter__(self) -> Iterator[np.ndarray]:
@@ -107,7 +107,7 @@ def _collate_and_decode(
         *,
         config: DatasetConfig,
         decoder: Optional[Callable[[io.IOBase], torch.Tensor]],
-    ):
+    ) -> Dict[str, Any]:
         image_array, label_array = data
 
         image: Union[torch.Tensor, io.BytesIO]
@@ -145,7 +145,7 @@ def _make_datapipe(
 
 class MNIST(_MNISTBase):
     @property
-    def info(self):
+    def info(self) -> DatasetInfo:
         return DatasetInfo(
             "mnist",
             type=DatasetType.RAW,
@@ -176,7 +176,7 @@ def _files_and_checksums(self, config: DatasetConfig) -> Tuple[Tuple[str, str],
 
 class FashionMNIST(MNIST):
     @property
-    def info(self):
+    def info(self) -> DatasetInfo:
         return DatasetInfo(
             "fashionmnist",
             type=DatasetType.RAW,
@@ -209,7 +209,7 @@ def info(self):
 
 class KMNIST(MNIST):
     @property
-    def info(self):
+    def info(self) -> DatasetInfo:
         return DatasetInfo(
             "kmnist",
             type=DatasetType.RAW,
@@ -231,7 +231,7 @@ def info(self):
 
 class EMNIST(_MNISTBase):
     @property
-    def info(self):
+    def info(self) -> DatasetInfo:
         return DatasetInfo(
             "emnist",
             type=DatasetType.RAW,
@@ -295,7 +295,7 @@ def _collate_and_decode(
         *,
         config: DatasetConfig,
         decoder: Optional[Callable[[io.IOBase], torch.Tensor]],
-    ):
+    ) -> Dict[str, Any]:
         image_array, label_array = data
         # In these two splits, some lowercase letters are merged into their uppercase ones (see Fig 2. in the paper).
         # That means for example that there is 'D', 'd', and 'C', but not 'c'. Since the labels are nevertheless dense,
@@ -321,7 +321,7 @@ def _make_datapipe(
         images_dp, labels_dp = Demultiplexer(
             archive_dp,
             2,
-            functools.partial(self._classify_archive, config=config),  # type:ignore[arg-type]
+            functools.partial(self._classify_archive, config=config),
             drop_none=True,
             buffer_size=INFINITE_BUFFER_SIZE,
         )
@@ -330,7 +330,7 @@ def _make_datapipe(
 
 class QMNIST(_MNISTBase):
     @property
-    def info(self):
+    def info(self) -> DatasetInfo:
         return DatasetInfo(
             "qmnist",
             type=DatasetType.RAW,
@@ -381,7 +381,7 @@ def _collate_and_decode(
         *,
         config: DatasetConfig,
         decoder: Optional[Callable[[io.IOBase], torch.Tensor]],
-    ):
+    ) -> Dict[str, Any]:
         image_array, label_array = data
         label_parts = label_array.tolist()
         sample = super()._collate_and_decode((image_array, label_parts[0]), config=config, decoder=decoder)
diff --git a/torchvision/prototype/datasets/_builtin/sbd.py b/torchvision/prototype/datasets/_builtin/sbd.py
index c0244aa534a..4654a20c61f 100644
--- a/torchvision/prototype/datasets/_builtin/sbd.py
+++ b/torchvision/prototype/datasets/_builtin/sbd.py
@@ -1,7 +1,7 @@
 import io
 import pathlib
 import re
-from typing import Any, Callable, Dict, List, Optional, Tuple
+from typing import Any, Callable, Dict, List, Optional, Tuple, cast
 
 import numpy as np
 import torch
@@ -135,7 +135,7 @@ def _make_datapipe(
         split_dp, images_dp, anns_dp = Demultiplexer(
             archive_dp,
             3,
-            self._classify_archive,  # type: ignore[arg-type]
+            self._classify_archive,
             buffer_size=INFINITE_BUFFER_SIZE,
             drop_none=True,
         )
@@ -165,9 +165,15 @@ def _generate_categories(self, root: pathlib.Path) -> Tuple[str, ...]:
         lines = tuple(zip(*iter(dp)))[1]
 
         pattern = re.compile(r"\s*'(?P<category>\w+)';\s*%(?P<label>\d+)")
-        categories_and_labels = [
-            pattern.match(line).groups()  # type: ignore[union-attr]
-            # the first and last line contain no information
-            for line in lines[1:-1]
-        ]
-        return tuple(zip(*sorted(categories_and_labels, key=lambda category_and_label: int(category_and_label[1]))))[0]
+        categories_and_labels = cast(
+            List[Tuple[str, str]],
+            [
+                pattern.match(line).groups()  # type: ignore[union-attr]
+                # the first and last line contain no information
+                for line in lines[1:-1]
+            ],
+        )
+        categories_and_labels.sort(key=lambda category_and_label: int(category_and_label[1]))
+        categories, _ = zip(*categories_and_labels)
+
+        return categories
diff --git a/torchvision/prototype/datasets/_builtin/voc.py b/torchvision/prototype/datasets/_builtin/voc.py
index 21872c6a876..4f51cc7d29d 100644
--- a/torchvision/prototype/datasets/_builtin/voc.py
+++ b/torchvision/prototype/datasets/_builtin/voc.py
@@ -92,7 +92,11 @@ def _decode_detection_ann(self, buffer: io.IOBase) -> torch.Tensor:
         return torch.tensor(bboxes)
 
     def _collate_and_decode_sample(
-        self, data, *, config: DatasetConfig, decoder: Optional[Callable[[io.IOBase], torch.Tensor]]
+        self,
+        data: Tuple[Tuple[Tuple[str, str], Tuple[str, io.IOBase]], Tuple[str, io.IOBase]],
+        *,
+        config: DatasetConfig,
+        decoder: Optional[Callable[[io.IOBase], torch.Tensor]],
     ) -> Dict[str, Any]:
         split_and_image_data, ann_data = data
         _, image_data = split_and_image_data
@@ -104,7 +108,7 @@ def _collate_and_decode_sample(
         if config.task == "detection":
             ann = self._decode_detection_ann(ann_buffer)
         else:  # config.task == "segmentation":
-            ann = decoder(ann_buffer) if decoder else ann_buffer
+            ann = decoder(ann_buffer) if decoder else ann_buffer  # type: ignore[assignment]
 
         return dict(image_path=image_path, image=image, ann_path=ann_path, ann=ann)
 
@@ -120,7 +124,7 @@ def _make_datapipe(
         split_dp, images_dp, anns_dp = Demultiplexer(
             archive_dp,
             3,
-            functools.partial(self._classify_archive, config=config),  # type: ignore[arg-type]
+            functools.partial(self._classify_archive, config=config),
             drop_none=True,
             buffer_size=INFINITE_BUFFER_SIZE,
         )
diff --git a/torchvision/prototype/datasets/benchmark.py b/torchvision/prototype/datasets/benchmark.py
index 39308371f48..b0fc0615ccc 100644
--- a/torchvision/prototype/datasets/benchmark.py
+++ b/torchvision/prototype/datasets/benchmark.py
@@ -1,3 +1,5 @@
+# type: ignore
+
 import argparse
 import collections.abc
 import contextlib
diff --git a/torchvision/prototype/datasets/generate_category_files.py b/torchvision/prototype/datasets/generate_category_files.py
index a5623cd6db2..32db4c894b1 100644
--- a/torchvision/prototype/datasets/generate_category_files.py
+++ b/torchvision/prototype/datasets/generate_category_files.py
@@ -1,3 +1,5 @@
+# type: ignore
+
 import argparse
 import sys
 import unittest.mock
@@ -58,7 +60,7 @@ def parse_args(argv=None):
 
 
 if __name__ == "__main__":
-    args = parse_args()
+    args = parse_args(["-f", "sbd"])
 
     try:
         main(*args.names, force=args.force)
diff --git a/torchvision/prototype/datasets/utils/_internal.py b/torchvision/prototype/datasets/utils/_internal.py
index 473851209a5..faceac74b6a 100644
--- a/torchvision/prototype/datasets/utils/_internal.py
+++ b/torchvision/prototype/datasets/utils/_internal.py
@@ -7,12 +7,12 @@
 import os.path
 import pathlib
 from typing import Collection, Sequence, Callable, Union, Any, Tuple, TypeVar, Iterator, Dict, Optional
+from typing import cast
 
 import numpy as np
 import PIL.Image
 from torch.utils.data import IterDataPipe
 
-
 __all__ = [
     "INFINITE_BUFFER_SIZE",
     "BUILTIN_DIR",
@@ -108,7 +108,7 @@ def __iter__(self) -> Iterator[Tuple[int, D]]:
 
 
 def getitem(*items: Any) -> Callable[[Any], Any]:
-    def wrapper(obj: Any):
+    def wrapper(obj: Any) -> Any:
         for item in items:
             obj = obj[item]
         return obj
@@ -121,7 +121,7 @@ def path_accessor(getter: Union[str, Callable[[pathlib.Path], D]]) -> Callable[[
         name = getter
 
         def getter(path: pathlib.Path) -> D:
-            return getattr(path, name)
+            return cast(D, getattr(path, name))
 
     def wrapper(data: Tuple[str, Any]) -> D:
         return getter(pathlib.Path(data[0]))  # type: ignore[operator]

From 7b1461822dcba53fa274e42b53f94344a8289d18 Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Wed, 20 Oct 2021 15:57:47 +0200
Subject: [PATCH 05/11] cleanup

---
 torchvision/prototype/datasets/_builtin/caltech.py |  8 ++++----
 torchvision/prototype/datasets/_builtin/celeba.py  |  8 ++++----
 torchvision/prototype/datasets/_builtin/cifar.py   | 10 +++++-----
 torchvision/prototype/datasets/_builtin/mnist.py   |  4 ++--
 torchvision/prototype/datasets/_builtin/sbd.py     |  4 ++--
 torchvision/prototype/datasets/_builtin/voc.py     |  6 ++----
 6 files changed, 19 insertions(+), 21 deletions(-)

diff --git a/torchvision/prototype/datasets/_builtin/caltech.py b/torchvision/prototype/datasets/_builtin/caltech.py
index 98fd70fe098..818ef3ec20c 100644
--- a/torchvision/prototype/datasets/_builtin/caltech.py
+++ b/torchvision/prototype/datasets/_builtin/caltech.py
@@ -120,11 +120,11 @@ def _make_datapipe(
         images_dp, anns_dp = resource_dps
 
         images_dp = TarArchiveReader(images_dp)
-        images_dp: IterDataPipe = Filter(images_dp, self._is_not_background_image)
+        images_dp = Filter(images_dp, self._is_not_background_image)
         images_dp = Shuffler(images_dp, buffer_size=INFINITE_BUFFER_SIZE)
 
         anns_dp = TarArchiveReader(anns_dp)
-        anns_dp: IterDataPipe = Filter(anns_dp, self._is_ann)
+        anns_dp = Filter(anns_dp, self._is_ann)
 
         dp = KeyZipper(
             images_dp,
@@ -139,7 +139,7 @@ def _make_datapipe(
     def _generate_categories(self, root: pathlib.Path) -> List[str]:
         dp = self.resources(self.default_config)[0].to_datapipe(pathlib.Path(root) / self.name)
         dp = TarArchiveReader(dp)
-        dp: IterDataPipe = Filter(dp, self._is_not_background_image)
+        dp = Filter(dp, self._is_not_background_image)
         return sorted({pathlib.Path(path).parent.name for path, _ in dp})
 
 
@@ -188,7 +188,7 @@ def _make_datapipe(
     ) -> IterDataPipe[Dict[str, Any]]:
         dp = resource_dps[0]
         dp = TarArchiveReader(dp)
-        dp: IterDataPipe = Filter(dp, self._is_not_rogue_file)
+        dp = Filter(dp, self._is_not_rogue_file)
         dp = Shuffler(dp, buffer_size=INFINITE_BUFFER_SIZE)
         return Mapper(dp, self._collate_and_decode_sample, fn_kwargs=dict(decoder=decoder))
 
diff --git a/torchvision/prototype/datasets/_builtin/celeba.py b/torchvision/prototype/datasets/_builtin/celeba.py
index 1a604cd756e..9ed1da4f24d 100644
--- a/torchvision/prototype/datasets/_builtin/celeba.py
+++ b/torchvision/prototype/datasets/_builtin/celeba.py
@@ -29,7 +29,7 @@
 class CelebACSVParser(IterDataPipe):
     def __init__(
         self,
-        datapipe: IterDataPipe[Tuple[str, io.IOBase]],
+        datapipe: IterDataPipe[Tuple[Any, io.IOBase]],
         *,
         has_header: bool,
     ) -> None:
@@ -156,12 +156,12 @@ def _make_datapipe(
         splits_dp, images_dp, identities_dp, attributes_dp, bboxes_dp, landmarks_dp = resource_dps
 
         splits_dp = CelebACSVParser(splits_dp, has_header=False)
-        splits_dp: IterDataPipe = Filter(splits_dp, self._filter_split, fn_kwargs=dict(split=config.split))
+        splits_dp = Filter(splits_dp, self._filter_split, fn_kwargs=dict(split=config.split))
         splits_dp = Shuffler(splits_dp, buffer_size=INFINITE_BUFFER_SIZE)
 
         images_dp = ZipArchiveReader(images_dp)
 
-        anns_dp: IterDataPipe = Zipper(
+        anns_dp = Zipper(
             *[
                 CelebACSVParser(dp, has_header=has_header)
                 for dp, has_header in (
@@ -172,7 +172,7 @@ def _make_datapipe(
                 )
             ]
         )
-        anns_dp: IterDataPipe = Mapper(anns_dp, self._collate_anns)
+        anns_dp = Mapper(anns_dp, self._collate_anns)
 
         dp = KeyZipper(
             splits_dp,
diff --git a/torchvision/prototype/datasets/_builtin/cifar.py b/torchvision/prototype/datasets/_builtin/cifar.py
index 4949c56f6cb..edb1c7f88d5 100644
--- a/torchvision/prototype/datasets/_builtin/cifar.py
+++ b/torchvision/prototype/datasets/_builtin/cifar.py
@@ -86,9 +86,9 @@ def _make_datapipe(
         decoder: Optional[Callable[[io.IOBase], torch.Tensor]],
     ) -> IterDataPipe[Dict[str, Any]]:
         dp = resource_dps[0]
-        dp: IterDataPipe = TarArchiveReader(dp)
-        dp: IterDataPipe = Filter(dp, functools.partial(self._is_data_file, config=config))
-        dp: IterDataPipe = Mapper(dp, self._unpickle)
+        dp = TarArchiveReader(dp)
+        dp = Filter(dp, functools.partial(self._is_data_file, config=config))
+        dp = Mapper(dp, self._unpickle)
         dp = CifarFileReader(dp, labels_key=self._LABELS_KEY)
         dp = Shuffler(dp, buffer_size=INFINITE_BUFFER_SIZE)
         return Mapper(dp, self._collate_and_decode, fn_kwargs=dict(decoder=decoder))
@@ -96,8 +96,8 @@ def _make_datapipe(
     def _generate_categories(self, root: pathlib.Path) -> List[str]:
         dp = self.resources(self.default_config)[0].to_datapipe(pathlib.Path(root) / self.name)
         dp = TarArchiveReader(dp)
-        dp: IterDataPipe = Filter(dp, path_comparator("name", self._META_FILE_NAME))
-        dp: IterDataPipe = Mapper(dp, self._unpickle)
+        dp = Filter(dp, path_comparator("name", self._META_FILE_NAME))
+        dp = Mapper(dp, self._unpickle)
         return cast(List[str], next(iter(dp))[self._CATEGORIES_KEY])
 
 
diff --git a/torchvision/prototype/datasets/_builtin/mnist.py b/torchvision/prototype/datasets/_builtin/mnist.py
index 9778c920f37..940a323fa19 100644
--- a/torchvision/prototype/datasets/_builtin/mnist.py
+++ b/torchvision/prototype/datasets/_builtin/mnist.py
@@ -48,7 +48,7 @@ class MNISTFileReader(IterDataPipe):
         14: "f8",  # float64
     }
 
-    def __init__(self, datapipe: IterDataPipe, *, start: Optional[int], stop: Optional[int]) -> None:
+    def __init__(self, datapipe: IterDataPipe[Any, io.IOBase], *, start: Optional[int], stop: Optional[int]) -> None:
         self.datapipe = datapipe
         self.start = start
         self.stop = stop
@@ -138,7 +138,7 @@ def _make_datapipe(
         labels_dp = Decompressor(labels_dp)
         labels_dp = MNISTFileReader(labels_dp, start=start, stop=stop)
 
-        dp: IterDataPipe = Zipper(images_dp, labels_dp)
+        dp = Zipper(images_dp, labels_dp)
         dp = Shuffler(dp, buffer_size=INFINITE_BUFFER_SIZE)
         return Mapper(dp, self._collate_and_decode, fn_kwargs=dict(config=config, decoder=decoder))
 
diff --git a/torchvision/prototype/datasets/_builtin/sbd.py b/torchvision/prototype/datasets/_builtin/sbd.py
index 4654a20c61f..b5f9a710288 100644
--- a/torchvision/prototype/datasets/_builtin/sbd.py
+++ b/torchvision/prototype/datasets/_builtin/sbd.py
@@ -159,9 +159,9 @@ def _make_datapipe(
     def _generate_categories(self, root: pathlib.Path) -> Tuple[str, ...]:
         dp = self.resources(self.default_config)[0].to_datapipe(pathlib.Path(root) / self.name)
         dp = TarArchiveReader(dp)
-        dp: IterDataPipe = Filter(dp, path_comparator("name", "category_names.m"))
+        dp = Filter(dp, path_comparator("name", "category_names.m"))
         dp = LineReader(dp)
-        dp: IterDataPipe = Mapper(dp, bytes.decode, input_col=1)
+        dp = Mapper(dp, bytes.decode, input_col=1)
         lines = tuple(zip(*iter(dp)))[1]
 
         pattern = re.compile(r"\s*'(?P<category>\w+)';\s*%(?P<label>\d+)")
diff --git a/torchvision/prototype/datasets/_builtin/voc.py b/torchvision/prototype/datasets/_builtin/voc.py
index 4f51cc7d29d..d4424d532dd 100644
--- a/torchvision/prototype/datasets/_builtin/voc.py
+++ b/torchvision/prototype/datasets/_builtin/voc.py
@@ -129,10 +129,8 @@ def _make_datapipe(
             buffer_size=INFINITE_BUFFER_SIZE,
         )
 
-        split_dp: IterDataPipe = Filter(
-            split_dp, self._is_in_folder, fn_kwargs=dict(name=self._SPLIT_FOLDER[config.task])
-        )
-        split_dp: IterDataPipe = Filter(split_dp, path_comparator("name", f"{config.split}.txt"))
+        split_dp = Filter(split_dp, self._is_in_folder, fn_kwargs=dict(name=self._SPLIT_FOLDER[config.task]))
+        split_dp = Filter(split_dp, path_comparator("name", f"{config.split}.txt"))
         split_dp = LineReader(split_dp, decode=True)
         split_dp = Shuffler(split_dp, buffer_size=INFINITE_BUFFER_SIZE)
 

From f7943d7e63eb6570d25155b6e4c5a9694dd6f2f9 Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Thu, 21 Oct 2021 12:14:43 +0200
Subject: [PATCH 06/11] fix mnist annotations

---
 torchvision/prototype/datasets/_builtin/mnist.py | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/torchvision/prototype/datasets/_builtin/mnist.py b/torchvision/prototype/datasets/_builtin/mnist.py
index 940a323fa19..af22199ce39 100644
--- a/torchvision/prototype/datasets/_builtin/mnist.py
+++ b/torchvision/prototype/datasets/_builtin/mnist.py
@@ -38,7 +38,7 @@
 prod = functools.partial(functools.reduce, operator.mul)
 
 
-class MNISTFileReader(IterDataPipe):
+class MNISTFileReader(IterDataPipe[np.ndarray]):
     _DTYPE_MAP = {
         8: "u1",  # uint8
         9: "i1",  # int8
@@ -48,7 +48,9 @@ class MNISTFileReader(IterDataPipe):
         14: "f8",  # float64
     }
 
-    def __init__(self, datapipe: IterDataPipe[Any, io.IOBase], *, start: Optional[int], stop: Optional[int]) -> None:
+    def __init__(
+        self, datapipe: IterDataPipe[Tuple[Any, io.IOBase]], *, start: Optional[int], stop: Optional[int]
+    ) -> None:
         self.datapipe = datapipe
         self.start = start
         self.stop = stop

From 735e41ff232d0e218834ce3d359b52bae6ef0c8c Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Thu, 21 Oct 2021 12:15:02 +0200
Subject: [PATCH 07/11] refactor celeba

---
 .../prototype/datasets/_builtin/celeba.py     | 46 +++++++++----------
 1 file changed, 21 insertions(+), 25 deletions(-)

diff --git a/torchvision/prototype/datasets/_builtin/celeba.py b/torchvision/prototype/datasets/_builtin/celeba.py
index 9ed1da4f24d..ebfce4b652d 100644
--- a/torchvision/prototype/datasets/_builtin/celeba.py
+++ b/torchvision/prototype/datasets/_builtin/celeba.py
@@ -1,6 +1,6 @@
 import csv
 import io
-from typing import Any, Callable, Dict, List, Optional, Tuple, Mapping, Union, Iterator
+from typing import Any, Callable, Dict, List, Optional, Tuple, Iterator, Sequence
 
 import torch
 from torchdata.datapipes.iter import (
@@ -26,21 +26,23 @@
 csv.register_dialect("celeba", delimiter=" ", skipinitialspace=True)
 
 
-class CelebACSVParser(IterDataPipe):
+class CelebACSVParser(IterDataPipe[Tuple[str, Dict[str, str]]]):
     def __init__(
         self,
         datapipe: IterDataPipe[Tuple[Any, io.IOBase]],
         *,
-        has_header: bool,
+        fieldnames: Optional[Sequence[str]] = None,
     ) -> None:
         self.datapipe = datapipe
-        self.has_header = has_header
+        self.fieldnames = fieldnames
 
-    def __iter__(self) -> Iterator[Tuple[str, Union[Dict[str, str], List[str]]]]:
+    def __iter__(self) -> Iterator[Tuple[str, Dict[str, str]]]:
         for _, file in self.datapipe:
             file = (line.decode() for line in file)
 
-            if self.has_header:
+            if self.fieldnames:
+                fieldnames = self.fieldnames
+            else:
                 # The first row is skipped, because it only contains the number of samples
                 next(file)
 
@@ -51,11 +53,8 @@ def __iter__(self) -> Iterator[Tuple[str, Union[Dict[str, str], List[str]]]]:
                 if fieldnames[0] != "image_id":
                     fieldnames.insert(0, "image_id")
 
-                for line_dict in csv.DictReader(file, fieldnames=fieldnames, dialect="celeba"):
-                    yield line_dict.pop("image_id"), line_dict
-            else:
-                for line_list in csv.reader(file, dialect="celeba"):
-                    yield line_list[0], line_list[1:]
+            for line in csv.DictReader(file, fieldnames=fieldnames, dialect="celeba"):
+                yield line.pop("image_id"), line
 
 
 class CelebA(Dataset):
@@ -106,13 +105,10 @@ def resources(self, config: DatasetConfig) -> List[OnlineResource]:
         "2": "test",
     }
 
-    def _filter_split(self, data: Tuple[str, str], *, split: str) -> bool:
-        _, split_id = data
-        return self._SPLIT_ID_TO_NAME[split_id[0]] == split
+    def _filter_split(self, data: Tuple[str, Dict[str, str]], *, split: str) -> bool:
+        return self._SPLIT_ID_TO_NAME[data[1]["split_id"]] == split
 
-    def _collate_anns(
-        self, data: Tuple[Tuple[str, Union[List[str], Mapping[str, str]]], ...]
-    ) -> Tuple[str, Dict[str, Union[List[str], Mapping[str, str]]]]:
+    def _collate_anns(self, data: Tuple[Tuple[str, Dict[str, str]], ...]) -> Tuple[str, Dict[str, Dict[str, str]]]:
         (image_id, identity), (_, attributes), (_, bbox), (_, landmarks) = data
         return image_id, dict(identity=identity, attributes=attributes, bbox=bbox, landmarks=landmarks)
 
@@ -129,7 +125,7 @@ def _collate_and_decode_sample(
 
         image = decoder(buffer) if decoder else buffer
 
-        identity = torch.tensor(int(ann["identity"][0]))
+        identity = int(ann["identity"]["identity"])
         attributes = {attr: value == "1" for attr, value in ann["attributes"].items()}
         bbox = torch.tensor([int(ann["bbox"][key]) for key in ("x_1", "y_1", "width", "height")])
         landmarks = {
@@ -155,7 +151,7 @@ def _make_datapipe(
     ) -> IterDataPipe[Dict[str, Any]]:
         splits_dp, images_dp, identities_dp, attributes_dp, bboxes_dp, landmarks_dp = resource_dps
 
-        splits_dp = CelebACSVParser(splits_dp, has_header=False)
+        splits_dp = CelebACSVParser(splits_dp, fieldnames=("image_id", "split_id"))
         splits_dp = Filter(splits_dp, self._filter_split, fn_kwargs=dict(split=config.split))
         splits_dp = Shuffler(splits_dp, buffer_size=INFINITE_BUFFER_SIZE)
 
@@ -163,12 +159,12 @@ def _make_datapipe(
 
         anns_dp = Zipper(
             *[
-                CelebACSVParser(dp, has_header=has_header)
-                for dp, has_header in (
-                    (identities_dp, False),
-                    (attributes_dp, True),
-                    (bboxes_dp, True),
-                    (landmarks_dp, True),
+                CelebACSVParser(dp, fieldnames=fieldnames)
+                for dp, fieldnames in (
+                    (identities_dp, ("image_id", "identity")),
+                    (attributes_dp, None),
+                    (bboxes_dp, None),
+                    (landmarks_dp, None),
                 )
             ]
         )

From 1c83855d067d7476dedbf9ca91716e28982effa3 Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Thu, 21 Oct 2021 12:25:46 +0200
Subject: [PATCH 08/11] warn on redundant casts

---
 mypy.ini | 1 +
 1 file changed, 1 insertion(+)

diff --git a/mypy.ini b/mypy.ini
index 6069eece9bc..d2bbe22614f 100644
--- a/mypy.ini
+++ b/mypy.ini
@@ -4,6 +4,7 @@ files = torchvision
 show_error_codes = True
 pretty = True
 allow_redefinition = True
+warn_redundant_casts = True
 
 [mypy-torchvision.prototype.datasets.*]
 

From 069d402d844156517f387172365237ce4a07c3b2 Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Thu, 21 Oct 2021 13:41:33 +0200
Subject: [PATCH 09/11] remove redundant cast

---
 torchvision/datasets/usps.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/torchvision/datasets/usps.py b/torchvision/datasets/usps.py
index c90ebfa7e6f..c09ec282e9d 100644
--- a/torchvision/datasets/usps.py
+++ b/torchvision/datasets/usps.py
@@ -1,5 +1,5 @@
 import os
-from typing import Any, Callable, cast, Optional, Tuple
+from typing import Any, Callable, Optional, Tuple
 
 import numpy as np
 from PIL import Image
@@ -63,7 +63,7 @@ def __init__(
             raw_data = [line.decode().split() for line in fp.readlines()]
             tmp_list = [[x.split(":")[-1] for x in data[1:]] for data in raw_data]
             imgs = np.asarray(tmp_list, dtype=np.float32).reshape((-1, 16, 16))
-            imgs = ((cast(np.ndarray, imgs) + 1) / 2 * 255).astype(dtype=np.uint8)
+            imgs = ((imgs + 1) / 2 * 255).astype(dtype=np.uint8)
             targets = [int(d[0]) - 1 for d in raw_data]
 
         self.data = imgs

From 2095cdb0cb3e45d91aaaadf5df33be5290bf1213 Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Thu, 21 Oct 2021 13:41:51 +0200
Subject: [PATCH 10/11] simplify annotation

---
 torchvision/prototype/datasets/_builtin/imagenet.py | 4 ++--
 torchvision/prototype/datasets/_builtin/sbd.py      | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/torchvision/prototype/datasets/_builtin/imagenet.py b/torchvision/prototype/datasets/_builtin/imagenet.py
index 56352249d5f..28b52d6985b 100644
--- a/torchvision/prototype/datasets/_builtin/imagenet.py
+++ b/torchvision/prototype/datasets/_builtin/imagenet.py
@@ -152,7 +152,7 @@ def _make_datapipe(
         "n03710721": "tank suit",
     }
 
-    def _generate_categories(self, root: pathlib.Path) -> List[Tuple[str, str]]:
+    def _generate_categories(self, root: pathlib.Path) -> List[Tuple[str, ...]]:
         resources = self.resources(self.default_config)
         devkit_dp = resources[1].to_datapipe(root / self.name)
         devkit_dp = TarArchiveReader(devkit_dp)
@@ -161,7 +161,7 @@ def _generate_categories(self, root: pathlib.Path) -> List[Tuple[str, str]]:
         meta = next(iter(devkit_dp))[1]
         synsets = read_mat(meta, squeeze_me=True)["synsets"]
         categories_and_wnids = cast(
-            List[Tuple[str, str]],
+            List[Tuple[str, ...]],
             [
                 (self._WNID_MAP.get(wnid, category.split(",", 1)[0]), wnid)
                 for _, wnid, category, _, num_children, *_ in synsets
diff --git a/torchvision/prototype/datasets/_builtin/sbd.py b/torchvision/prototype/datasets/_builtin/sbd.py
index b5f9a710288..3f57f488795 100644
--- a/torchvision/prototype/datasets/_builtin/sbd.py
+++ b/torchvision/prototype/datasets/_builtin/sbd.py
@@ -166,7 +166,7 @@ def _generate_categories(self, root: pathlib.Path) -> Tuple[str, ...]:
 
         pattern = re.compile(r"\s*'(?P<category>\w+)';\s*%(?P<label>\d+)")
         categories_and_labels = cast(
-            List[Tuple[str, str]],
+            List[Tuple[str, ...]],
             [
                 pattern.match(line).groups()  # type: ignore[union-attr]
                 # the first and last line contain no information

From 0a954d300a14362e7b5dbeec9fd5b7939ed9f2bc Mon Sep 17 00:00:00 2001
From: Philip Meier <github.pmeier@posteo.de>
Date: Thu, 21 Oct 2021 13:44:32 +0200
Subject: [PATCH 11/11] fix import

---
 torchvision/prototype/datasets/utils/_internal.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/torchvision/prototype/datasets/utils/_internal.py b/torchvision/prototype/datasets/utils/_internal.py
index 3dae8c6b40f..fdb6a17c46f 100644
--- a/torchvision/prototype/datasets/utils/_internal.py
+++ b/torchvision/prototype/datasets/utils/_internal.py
@@ -101,7 +101,7 @@ def to_str(sep: str) -> str:
     return f"{prefix}\n{body}\n{postfix}"
 
 
-class FrozenMapping(collections.abc.Mapping[K, D]):
+class FrozenMapping(Mapping[K, D]):
     def __init__(self, *args: Any, **kwargs: Any) -> None:
         data = dict(*args, **kwargs)
         self.__dict__["__data__"] = data