Merge branch 'main' into replace

Author: NicolasHug

.circleci/unittest/linux/scripts/install.sh

@@ -21,7 +21,7 @@ else
     fi
     echo "Using CUDA $CUDA_VERSION as determined by CU_VERSION"
     version="$(python -c "print('.'.join(\"${CUDA_VERSION}\".split('.')[:2]))")"
-    cudatoolkit="cudatoolkit=${version}"
+    cudatoolkit="nvidia::cudatoolkit=${version}"
 fi
 
 case "$(uname -s)" in
@@ -33,7 +33,7 @@ printf "Installing PyTorch with %s\n" "${cudatoolkit}"
 if [ "${os}" == "MacOSX" ]; then
     conda install -y -c "pytorch-${UPLOAD_CHANNEL}" "pytorch-${UPLOAD_CHANNEL}"::pytorch "${cudatoolkit}" pytest
 else
-    conda install -y -c "pytorch-${UPLOAD_CHANNEL}" "pytorch-${UPLOAD_CHANNEL}"::pytorch[build="*${version}*"] "${cudatoolkit}" pytest
+    conda install -y -c "pytorch-${UPLOAD_CHANNEL}" -c nvidia "pytorch-${UPLOAD_CHANNEL}"::pytorch[build="*${version}*"] "${cudatoolkit}" pytest
 fi
 
 printf "* Installing torchvision\n"

packaging/build_cmake.sh

@@ -42,7 +42,7 @@ else
   PYTORCH_MUTEX_CONSTRAINT=''
 fi
 
-conda install -yq \pytorch=$PYTORCH_VERSION $CONDA_CUDATOOLKIT_CONSTRAINT $PYTORCH_MUTEX_CONSTRAINT $MKL_CONSTRAINT numpy -c "pytorch-${UPLOAD_CHANNEL}"
+conda install -yq \pytorch=$PYTORCH_VERSION $CONDA_CUDATOOLKIT_CONSTRAINT $PYTORCH_MUTEX_CONSTRAINT $MKL_CONSTRAINT numpy -c nvidia -c "pytorch-${UPLOAD_CHANNEL}"
 TORCH_PATH=$(dirname $(python -c "import torch; print(torch.__file__)"))
 
 if [[ "$(uname)" == Darwin || "$OSTYPE" == "msys" ]]; then

packaging/build_conda.sh

@@ -18,4 +18,4 @@ if [[ "$CU_VERSION" == cu115 ]]; then
     export CUDATOOLKIT_CHANNEL="conda-forge"
 fi
 
-conda build -c defaults -c $CUDATOOLKIT_CHANNEL $CONDA_CHANNEL_FLAGS --no-anaconda-upload --python "$PYTHON_VERSION" packaging/torchvision
+conda build -c $CUDATOOLKIT_CHANNEL -c defaults $CONDA_CHANNEL_FLAGS --no-anaconda-upload --python "$PYTHON_VERSION" packaging/torchvision

test/test_image.py

@@ -158,7 +158,7 @@ def test_decode_png(img_path, pil_mode, mode):
 
     img_pil = normalize_dimensions(img_pil)
 
-    if "16" in img_path:
+    if img_path.endswith("16.png"):
         # 16 bits image decoding is supported, but only as a private API
         # FIXME: see https://github.com/pytorch/vision/issues/4731 for potential solutions to making it public
         with pytest.raises(RuntimeError, match="At most 8-bit PNG images are supported"):

test/test_models.py

@@ -406,7 +406,7 @@ def test_mobilenet_norm_layer(model_fn):
     assert any(isinstance(x, nn.BatchNorm2d) for x in model.modules())
 
     def get_gn(num_channels):
-        return nn.GroupNorm(32, num_channels)
+        return nn.GroupNorm(1, num_channels)
 
     model = model_fn(norm_layer=get_gn)
     assert not (any(isinstance(x, nn.BatchNorm2d) for x in model.modules()))

torchvision/prototype/transforms/_augment.py

@@ -7,18 +7,19 @@
 from torchvision.prototype import features
 from torchvision.prototype.transforms import Transform, functional as F
 
+from ._transform import _RandomApplyTransform
 from ._utils import query_image, get_image_dimensions, has_all, has_any, is_simple_tensor
 
 
-class RandomErasing(Transform):
+class RandomErasing(_RandomApplyTransform):
     def __init__(
         self,
         p: float = 0.5,
         scale: Tuple[float, float] = (0.02, 0.33),
         ratio: Tuple[float, float] = (0.3, 3.3),
         value: float = 0,
     ):
-        super().__init__()
+        super().__init__(p=p)
         if not isinstance(value, (numbers.Number, str, tuple, list)):
             raise TypeError("Argument value should be either a number or str or a sequence")
         if isinstance(value, str) and value != "random":
@@ -31,9 +32,6 @@ def __init__(
             warnings.warn("Scale and ratio should be of kind (min, max)")
         if scale[0] < 0 or scale[1] > 1:
             raise ValueError("Scale should be between 0 and 1")
-        if p < 0 or p > 1:
-            raise ValueError("Random erasing probability should be between 0 and 1")
-        self.p = p
         self.scale = scale
         self.ratio = ratio
         self.value = value
@@ -99,8 +97,6 @@ def forward(self, *inputs: Any) -> Any:
         sample = inputs if len(inputs) > 1 else inputs[0]
         if has_any(sample, features.BoundingBox, features.SegmentationMask):
             raise TypeError(f"BoundingBox'es and SegmentationMask's are not supported by {type(self).__name__}()")
-        elif torch.rand(1) >= self.p:
-            return sample
 
         return super().forward(sample)
 

torchvision/prototype/transforms/_container.py

@@ -1,8 +1,9 @@
-from typing import Any, Optional, List
+from typing import Any, Optional, List, Dict
 
 import torch
+from torchvision.prototype.transforms import Transform
 
-from ._transform import Transform
+from ._transform import _RandomApplyTransform
 
 
 class Compose(Transform):
@@ -19,18 +20,13 @@ def forward(self, *inputs: Any) -> Any:
         return sample
 
 
-class RandomApply(Transform):
+class RandomApply(_RandomApplyTransform):
     def __init__(self, transform: Transform, *, p: float = 0.5) -> None:
-        super().__init__()
+        super().__init__(p=p)
         self.transform = transform
-        self.p = p
-
-    def forward(self, *inputs: Any) -> Any:
-        sample = inputs if len(inputs) > 1 else inputs[0]
-        if float(torch.rand(())) < self.p:
-            return sample
 
-        return self.transform(sample)
+    def _transform(self, input: Any, params: Dict[str, Any]) -> Any:
+        return self.transform(input)
 
     def extra_repr(self) -> str:
         return f"p={self.p}"

torchvision/prototype/transforms/_geometry.py

@@ -12,21 +12,11 @@
 from torchvision.transforms.transforms import _setup_size, _interpolation_modes_from_int
 from typing_extensions import Literal
 
+from ._transform import _RandomApplyTransform
 from ._utils import query_image, get_image_dimensions, has_any, is_simple_tensor
 
 
-class RandomHorizontalFlip(Transform):
-    def __init__(self, p: float = 0.5) -> None:
-        super().__init__()
-        self.p = p
-
-    def forward(self, *inputs: Any) -> Any:
-        sample = inputs if len(inputs) > 1 else inputs[0]
-        if torch.rand(1) >= self.p:
-            return sample
-
-        return super().forward(sample)
-
+class RandomHorizontalFlip(_RandomApplyTransform):
     def _transform(self, input: Any, params: Dict[str, Any]) -> Any:
         if isinstance(input, features.Image):
             output = F.horizontal_flip_image_tensor(input)
@@ -45,18 +35,7 @@ def _transform(self, input: Any, params: Dict[str, Any]) -> Any:
             return input
 
 
-class RandomVerticalFlip(Transform):
-    def __init__(self, p: float = 0.5) -> None:
-        super().__init__()
-        self.p = p
-
-    def forward(self, *inputs: Any) -> Any:
-        sample = inputs if len(inputs) > 1 else inputs[0]
-        if torch.rand(1) > self.p:
-            return sample
-
-        return super().forward(sample)
-
+class RandomVerticalFlip(_RandomApplyTransform):
     def _transform(self, input: Any, params: Dict[str, Any]) -> Any:
         if isinstance(input, features.Image):
             output = F.vertical_flip_image_tensor(input)
@@ -371,11 +350,11 @@ def _transform(self, input: Any, params: Dict[str, Any]) -> Any:
             return input
 
 
-class RandomZoomOut(Transform):
+class RandomZoomOut(_RandomApplyTransform):
     def __init__(
         self, fill: Union[float, Sequence[float]] = 0.0, side_range: Tuple[float, float] = (1.0, 4.0), p: float = 0.5
     ) -> None:
-        super().__init__()
+        super().__init__(p=p)
 
         if fill is None:
             fill = 0.0
@@ -385,8 +364,6 @@ def __init__(
         if side_range[0] < 1.0 or side_range[0] > side_range[1]:
             raise ValueError(f"Invalid canvas side range provided {side_range}.")
 
-        self.p = p
-
     def _get_params(self, sample: Any) -> Dict[str, Any]:
         image = query_image(sample)
         orig_c, orig_h, orig_w = get_image_dimensions(image)
@@ -411,10 +388,3 @@ def _get_params(self, sample: Any) -> Dict[str, Any]:
     def _transform(self, input: Any, params: Dict[str, Any]) -> Any:
         transform = Pad(**params, padding_mode="constant")
         return transform(input)
-
-    def forward(self, *inputs: Any) -> Any:
-        sample = inputs if len(inputs) > 1 else inputs[0]
-        if torch.rand(1) >= self.p:
-            return sample
-
-        return super().forward(sample)

torchvision/prototype/transforms/_transform.py

@@ -2,6 +2,7 @@
 import functools
 from typing import Any, Dict
 
+import torch
 from torch import nn
 from torchvision.prototype.utils._internal import apply_recursively
 from torchvision.utils import _log_api_usage_once
@@ -34,3 +35,20 @@ def extra_repr(self) -> str:
             extra.append(f"{name}={value}")
 
         return ", ".join(extra)
+
+
+class _RandomApplyTransform(Transform):
+    def __init__(self, *, p: float = 0.5) -> None:
+        if not (0.0 <= p <= 1.0):
+            raise ValueError("`p` should be a floating point value in the interval [0.0, 1.0].")
+
+        super().__init__()
+        self.p = p
+
+    def forward(self, *inputs: Any) -> Any:
+        sample = inputs if len(inputs) > 1 else inputs[0]
+
+        if torch.rand(1) >= self.p:
+            return sample
+
+        return super().forward(sample)