Issue 2350 - support of all padding modes with tensors (#2368)

* [WIP] functional_tensor supports more padding modes

* [WIP] Support all padding modes

* Removed wip symmetric mode

* Improvements according to the review

Author: vfdev-5

test/test_functional_tensor.py

@@ -247,18 +247,36 @@ def test_ten_crop(self):
     def test_pad(self):
         script_fn = torch.jit.script(F_t.pad)
         tensor, pil_img = self._create_data(7, 8)
-        for pad in [1, [1, ], [0, 1], (2, 2), [1, 0, 1, 2]]:
-            padding_mode = "constant"
-            for fill in [0, 10, 20]:
-                pad_tensor = F_t.pad(tensor, pad, fill=fill, padding_mode=padding_mode)
-                pad_pil_img = F_pil.pad(pil_img, pad, fill=fill, padding_mode=padding_mode)
-                self.compareTensorToPIL(pad_tensor, pad_pil_img, msg="{}, {}".format(pad, fill))
-                if isinstance(pad, int):
-                    script_pad = [pad, ]
-                else:
-                    script_pad = pad
-                pad_tensor_script = script_fn(tensor, script_pad, fill=fill, padding_mode=padding_mode)
-                self.assertTrue(pad_tensor.equal(pad_tensor_script), msg="{}, {}".format(pad, fill))
+
+        for dt in [None, torch.float32, torch.float64]:
+            if dt is not None:
+                # This is a trivial cast to float of uint8 data to test all cases
+                tensor = tensor.to(dt)
+            for pad in [2, [3, ], [0, 3], (3, 3), [4, 2, 4, 3]]:
+                configs = [
+                    {"padding_mode": "constant", "fill": 0},
+                    {"padding_mode": "constant", "fill": 10},
+                    {"padding_mode": "constant", "fill": 20},
+                    {"padding_mode": "edge"},
+                    {"padding_mode": "reflect"},
+                ]
+                for kwargs in configs:
+                    pad_tensor = F_t.pad(tensor, pad, **kwargs)
+                    pad_pil_img = F_pil.pad(pil_img, pad, **kwargs)
+
+                    pad_tensor_8b = pad_tensor
+                    # we need to cast to uint8 to compare with PIL image
+                    if pad_tensor_8b.dtype != torch.uint8:
+                        pad_tensor_8b = pad_tensor_8b.to(torch.uint8)
+
+                    self.compareTensorToPIL(pad_tensor_8b, pad_pil_img, msg="{}, {}".format(pad, kwargs))
+
+                    if isinstance(pad, int):
+                        script_pad = [pad, ]
+                    else:
+                        script_pad = pad
+                    pad_tensor_script = script_fn(tensor, script_pad, **kwargs)
+                    self.assertTrue(pad_tensor.equal(pad_tensor_script), msg="{}, {}".format(pad, kwargs))
 
 
 if __name__ == '__main__':

torchvision/transforms/functional_tensor.py

@@ -346,7 +346,7 @@ def _hsv2rgb(img):
     return torch.einsum("ijk, xijk -> xjk", mask.to(dtype=img.dtype), a4)
 
 
-def pad(img: Tensor, padding: List[int], fill: int, padding_mode: str = "constant") -> Tensor:
+def pad(img: Tensor, padding: List[int], fill: int = 0, padding_mode: str = "constant") -> Tensor:
     r"""Pad the given Tensor Image on all sides with specified padding mode and fill value.
 
     Args:
@@ -363,6 +363,13 @@ def pad(img: Tensor, padding: List[int], fill: int, padding_mode: str = "constan
 
             - constant: pads with a constant value, this value is specified with fill
 
+            - edge: pads with the last value on the edge of the image
+
+            - reflect: pads with reflection of image (without repeating the last value on the edge)
+
+                       padding [1, 2, 3, 4] with 2 elements on both sides in reflect mode
+                       will result in [3, 2, 1, 2, 3, 4, 3, 2]
+
     Returns:
         Tensor: Padded image.
     """
@@ -383,8 +390,8 @@ def pad(img: Tensor, padding: List[int], fill: int, padding_mode: str = "constan
         raise ValueError("Padding must be an int or a 1, 2, or 4 element tuple, not a " +
                          "{} element tuple".format(len(padding)))
 
-    if padding_mode not in ["constant", ]:
-        raise ValueError("Only constant padding_mode supported for torch tensors")
+    if padding_mode not in ["constant", "edge", "reflect"]:
+        raise ValueError("Padding mode should be either constant, edge or reflect")
 
     if isinstance(padding, int):
         if torch.jit.is_scripting():
@@ -403,5 +410,30 @@ def pad(img: Tensor, padding: List[int], fill: int, padding_mode: str = "constan
 
     p = [pad_left, pad_right, pad_top, pad_bottom]
 
+    if padding_mode == "edge":
+        # remap padding_mode str
+        padding_mode = "replicate"
+
+    need_squeeze = False
+    if img.ndim < 4:
+        img = img.unsqueeze(dim=0)
+        need_squeeze = True
+
+    out_dtype = img.dtype
+    need_cast = False
+    if (padding_mode != "constant") and img.dtype not in (torch.float32, torch.float64):
+        # Here we temporary cast input tensor to float
+        # until pytorch issue is resolved :
+        # https://github.com/pytorch/pytorch/issues/40763
+        need_cast = True
+        img = img.to(torch.float32)
+
     img = torch.nn.functional.pad(img, p, mode=padding_mode, value=float(fill))
+
+    if need_squeeze:
+        img = img.squeeze(dim=0)
+
+    if need_cast:
+        img = img.to(out_dtype)
+
     return img