Docstring Fix for PILToTensor in Torchvision (pytorch#9254)

Summary:

pytorch#9221 identifies a confusion around image shape conventions for ToTensor and PILToTensor classes. The docstring has the following statement:
Converts a PIL Image (H x W x C) to a Tensor of shape (C x H x W).
This is confusing since PIL Image shape is not (H x W x C) but rather PIL Images expose their size as (W, H) via the size attribute, not as a shape tuple.
Proposed Docstring Update
Convert a PIL Image with H height, W width, and C channels to a Tensor of shape (C x H x W).

Reviewed By: AntoineSimoulin

Differential Revision: D85779518

Author: Zhitao Yu

torchvision/transforms/transforms.py

@@ -145,7 +145,15 @@ class PILToTensor:
 
     This transform does not support torchscript.
 
-    Converts a PIL Image (H x W x C) to a Tensor of shape (C x H x W).
+    Convert a PIL Image with H height, W width, and C channels to a Tensor of shape (C x H x W).
+
+    Example:
+        >>> from PIL import Image
+        >>> import torchvision.transforms as T
+        >>> img = Image.new("RGB", (320, 240))  # size (W=320, H=240)
+        >>> tensor = T.PILToTensor()(img)
+        >>> print(tensor.shape)
+        torch.Size([3, 240, 320])
     """
 
     def __init__(self) -> None:

torchvision/transforms/v2/_type_conversion.py

@@ -15,7 +15,15 @@ class PILToTensor(Transform):
 
     This transform does not support torchscript.
 
-    Converts a PIL Image (H x W x C) to a Tensor of shape (C x H x W).
+    Convert a PIL Image with H height, W width, and C channels to a Tensor of shape (C x H x W).
+
+    Example:
+        >>> from PIL import Image
+        >>> from torchvision.transforms import v2
+        >>> img = Image.new("RGB", (320, 240))  # size (W=320, H=240)
+        >>> tensor = v2.PILToTensor()(img)
+        >>> print(tensor.shape)
+        torch.Size([3, 240, 320])
     """
 
     _transformed_types = (PIL.Image.Image,)