[fbsync] Explicitly copying array in pil_to_tensor (#4566)

Summary:
* mExplicitly copying array in pil_to_tensor

* Update torchvision/transforms/functional.py

* Adding comments regarding implicit array deep copy in PILToTensor transform

* Update torchvision/transforms/transforms.py

Reviewed By: NicolasHug

Differential Revision: D31505570

fbshipit-source-id: a2ce1c5cf6b70236bc71a79a3d190ddac7d378a4

Co-authored-by: Nicolas Hug <[email protected]>
Co-authored-by: Nicolas Hug <[email protected]>
Co-authored-by: Nicolas Hug <[email protected]>
Co-authored-by: Vasilis Vryniotis <[email protected]>

Author: 3 people

torchvision/transforms/functional.py

@@ -156,6 +156,10 @@ def pil_to_tensor(pic):
 
     See :class:`~torchvision.transforms.PILToTensor` for more details.
 
+    .. note::
+
+        A deep copy of the underlying array is performed.
+
     Args:
         pic (PIL Image): Image to be converted to tensor.
 
@@ -172,7 +176,7 @@ def pil_to_tensor(pic):
         return torch.as_tensor(nppic)
 
     # handle PIL Image
-    img = torch.as_tensor(np.asarray(pic))
+    img = torch.as_tensor(np.array(pic, copy=True))
     img = img.view(pic.size[1], pic.size[0], len(pic.getbands()))
     # put it from HWC to CHW format
     img = img.permute((2, 0, 1))

torchvision/transforms/transforms.py

@@ -142,6 +142,10 @@ class PILToTensor:
 
     def __call__(self, pic):
         """
+        .. note::
+
+            A deep copy of the underlying array is performed.
+
         Args:
             pic (PIL Image): Image to be converted to tensor.