Support empty target_type for CelebA dataset (#1351)

* Support empty target_type for CelebA dataset

* Return (X, None) for interface consistency

* Document behavior for target_type=[]

* Simplify diff

* Raise exception on meaningless parameters

torchvision/datasets/celeba.py

@@ -21,7 +21,7 @@ class CelebA(VisionDataset):
                 ``bbox`` (np.array shape=(4,) dtype=int): bounding box (x, y, width, height)
                 ``landmarks`` (np.array shape=(10,) dtype=int): landmark points (lefteye_x, lefteye_y, righteye_x,
                     righteye_y, nose_x, nose_y, leftmouth_x, leftmouth_y, rightmouth_x, rightmouth_y)
-            Defaults to ``attr``.
+            Defaults to ``attr``. If empty, ``None`` will be returned as target.
         transform (callable, optional): A function/transform that  takes in an PIL image
             and returns a transformed version. E.g, ``transforms.ToTensor``
         target_transform (callable, optional): A function/transform that takes in the
@@ -59,6 +59,9 @@ class CelebA(VisionDataset):
         else:
             self.target_type = [target_type]
 
+        if not self.target_type and self.target_transform is not None:
+            raise RuntimeError('target_transform is specified but target_type is empty')
+
         if download:
             self.download()
 
@@ -133,13 +136,17 @@ class CelebA(VisionDataset):
             else:
                 # TODO: refactor with utils.verify_str_arg
                 raise ValueError("Target type \"{}\" is not recognized.".format(t))
-        target = tuple(target) if len(target) > 1 else target[0]
 
         if self.transform is not None:
             X = self.transform(X)
 
+        if target:
+            target = tuple(target) if len(target) > 1 else target[0]
+
             if self.target_transform is not None:
                 target = self.target_transform(target)
+        else:
+            target = None
 
         return X, target