Merge branch 'main' into rotated-bboxes-transforms

Author: AntoineSimoulin

docs/source/training_references.rst

@@ -19,9 +19,9 @@ guarantees.
 
 In general, these scripts rely on the latest (not yet released) pytorch version
 or the latest torchvision version. This means that to use them, **you might need
-to install the latest pytorch and torchvision versions**, with e.g.::
+to install the latest pytorch and torchvision versions** following the `official
+instructions <https://pytorch.org/get-started/locally/>`_.
 
-    conda install pytorch torchvision -c pytorch-nightly
 
 If you need to rely on an older stable version of pytorch or torchvision, e.g.
 torchvision 0.10, then it's safer to use the scripts from that corresponding

mypy.ini

@@ -29,6 +29,10 @@ ignore_errors = True
 
 ignore_errors = True
 
+[mypy-torchvision.prototype.models.*]
+
+ignore_errors = True
+
 [mypy-torchvision.io.image.*]
 
 ignore_errors = True
@@ -41,54 +45,10 @@ ignore_errors = True
 
 ignore_errors = True
 
-[mypy-torchvision.models.densenet.*]
+[mypy-torchvision.models.*]
 
 ignore_errors=True
 
-[mypy-torchvision.models.maxvit.*]
-
-ignore_errors=True
-
-[mypy-torchvision.models.detection.anchor_utils]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.transform]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.roi_heads]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.faster_rcnn]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.mask_rcnn]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.keypoint_rcnn]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.retinanet]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.ssd]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.ssdlite]
-
-ignore_errors = True
-
-[mypy-torchvision.models.detection.fcos]
-
-ignore_errors = True
-
 [mypy-torchvision.ops.*]
 
 ignore_errors = True

setup.py

@@ -24,7 +24,7 @@
 NVCC_FLAGS = os.getenv("NVCC_FLAGS", None)
 # Note: the GPU video decoding stuff used to be called "video codec", which
 # isn't an accurate or descriptive name considering there are at least 2 other
-# video deocding backends in torchvision. I'm renaming this to "gpu video
+# video decoding backends in torchvision. I'm renaming this to "gpu video
 # decoder" where possible, keeping user facing names (like the env var below) to
 # the old scheme for BC.
 USE_GPU_VIDEO_DECODER = os.getenv("TORCHVISION_USE_VIDEO_CODEC", "1") == "1"
@@ -211,7 +211,7 @@ def find_libpng():
             subprocess.run([libpng_config, "--version"], stdout=subprocess.PIPE).stdout.strip().decode("utf-8")
         )
         if png_version < min_version:
-            warnings.warn("libpng version {png_version} is less than minimum required version {min_version}")
+            warnings.warn(f"libpng version {png_version} is less than minimum required version {min_version}")
             return False, None, None, None
 
         include_dir = (
@@ -448,7 +448,7 @@ def find_ffmpeg_libraries():
 
         extensions.append(
             CppExtension(
-                # This is an aweful name. It should be "cpu_video_decoder". Keeping for BC.
+                # This is an awful name. It should be "cpu_video_decoder". Keeping for BC.
                 "torchvision.video_reader",
                 combined_src,
                 include_dirs=[

torchvision/datasets/lfw.py

@@ -51,6 +51,10 @@ def __init__(
         self.data: list[Any] = []
 
         if download:
+            raise ValueError(
+                "LFW dataset is no longer available for download."
+                "Please download the dataset manually and place it in the specified directory"
+            )
             self.download()
 
         if not self._check_integrity():
@@ -90,6 +94,11 @@ def __len__(self) -> int:
 class LFWPeople(_LFW):
     """`LFW <http://vis-www.cs.umass.edu/lfw/>`_ Dataset.
 
+    .. warning:
+
+        The LFW dataset is no longer available for automatic download. Please
+        download it manually and place it in the specified directory.
+
     Args:
         root (str or ``pathlib.Path``): Root directory of dataset where directory
             ``lfw-py`` exists or will be saved to if download is set to True.
@@ -101,9 +110,7 @@ class LFWPeople(_LFW):
             and returns a transformed version. E.g, ``transforms.RandomCrop``
         target_transform (callable, optional): A function/transform that takes in the
             target and transforms it.
-        download (bool, optional): If true, downloads the dataset from the internet and
-            puts it in root directory. If dataset is already downloaded, it is not
-            downloaded again.
+        download (bool, optional): NOT SUPPORTED ANYMORE, leave to False.
         loader (callable, optional): A function to load an image given its path.
             By default, it uses PIL as its image loader, but users could also pass in
             ``torchvision.io.decode_image`` for decoding image data into tensors directly.
@@ -175,6 +182,11 @@ def extra_repr(self) -> str:
 class LFWPairs(_LFW):
     """`LFW <http://vis-www.cs.umass.edu/lfw/>`_ Dataset.
 
+    .. warning:
+
+        The LFW dataset is no longer available for automatic download. Please
+        download it manually and place it in the specified directory.
+
     Args:
         root (str or ``pathlib.Path``): Root directory of dataset where directory
             ``lfw-py`` exists or will be saved to if download is set to True.
@@ -186,9 +198,7 @@ class LFWPairs(_LFW):
             and returns a transformed version. E.g, ``transforms.RandomRotation``
         target_transform (callable, optional): A function/transform that takes in the
             target and transforms it.
-        download (bool, optional): If true, downloads the dataset from the internet and
-            puts it in root directory. If dataset is already downloaded, it is not
-            downloaded again.
+        download (bool, optional): NOT SUPPORTED ANYMORE, leave to False.
         loader (callable, optional): A function to load an image given its path.
             By default, it uses PIL as its image loader, but users could also pass in
             ``torchvision.io.decode_image`` for decoding image data into tensors directly.

torchvision/models/detection/anchor_utils.py

@@ -74,20 +74,16 @@ def generate_anchors(
         return base_anchors.round()
 
     def set_cell_anchors(self, dtype: torch.dtype, device: torch.device):
-        return [cell_anchor.to(dtype=dtype, device=device) for cell_anchor in self.cell_anchors]
+        self.cell_anchors = [cell_anchor.to(dtype=dtype, device=device) for cell_anchor in self.cell_anchors]
 
     def num_anchors_per_location(self) -> list[int]:
         return [len(s) * len(a) for s, a in zip(self.sizes, self.aspect_ratios)]
 
     # For every combination of (a, (g, s), i) in (self.cell_anchors, zip(grid_sizes, strides), 0:2),
     # output g[i] anchors that are s[i] distance apart in direction i, with the same dimensions as a.
-    def grid_anchors(
-        self,
-        grid_sizes: list[list[int]],
-        strides: list[list[Tensor]],
-        cell_anchors: list[torch.Tensor],
-    ) -> list[Tensor]:
+    def grid_anchors(self, grid_sizes: list[list[int]], strides: list[list[Tensor]]) -> list[Tensor]:
         anchors = []
+        cell_anchors = self.cell_anchors
         torch._assert(cell_anchors is not None, "cell_anchors should not be None")
         torch._assert(
             len(grid_sizes) == len(strides) == len(cell_anchors),
@@ -127,8 +123,8 @@ def forward(self, image_list: ImageList, feature_maps: list[Tensor]) -> list[Ten
             ]
             for g in grid_sizes
         ]
-        cell_anchors = self.set_cell_anchors(dtype, device)
-        anchors_over_all_feature_maps = self.grid_anchors(grid_sizes, strides, cell_anchors)
+        self.set_cell_anchors(dtype, device)
+        anchors_over_all_feature_maps = self.grid_anchors(grid_sizes, strides)
         anchors: list[list[torch.Tensor]] = []
         for _ in range(len(image_list.image_sizes)):
             anchors_in_image = [anchors_per_feature_map for anchors_per_feature_map in anchors_over_all_feature_maps]

torchvision/models/detection/mask_rcnn.py

@@ -50,7 +50,7 @@ class MaskRCNN(FasterRCNN):
           ``0 <= x1 < x2 <= W`` and ``0 <= y1 < y2 <= H``.
         - labels (Int64Tensor[N]): the predicted labels for each image
         - scores (Tensor[N]): the scores or each prediction
-        - masks (UInt8Tensor[N, 1, H, W]): the predicted masks for each instance, in 0-1 range. In order to
+        - masks (FloatTensor[N, 1, H, W]): the predicted masks for each instance, in 0-1 range. In order to
           obtain the final segmentation masks, the soft masks can be thresholded, generally
           with a value of 0.5 (mask >= 0.5)
 

torchvision/transforms/v2/functional/_geometry.py

@@ -1099,7 +1099,7 @@ def _rotate_image_pil(
     interpolation = _check_interpolation(interpolation)
 
     return _FP.rotate(
-        image, angle, interpolation=pil_modes_mapping[interpolation], expand=expand, fill=fill, center=center
+        image, angle, interpolation=pil_modes_mapping[interpolation], expand=expand, fill=fill, center=center  # type: ignore[arg-type]
     )
 
 

version.txt

@@ -1 +1 @@
-0.22.0a0
+0.23.0a0