PIL Image.transform should take in size (width, height)

image/randaugment/augmentation_transforms.py

@@ -49,6 +49,11 @@ def get_mean_and_std():
   return means, stds
 
 
+def _width_height_from_img_shape(img_shape):
+  """`img_shape` in autoaugment is (height, width)."""
+  return (img_shape[1], img_shape[0])
+
+
 def random_flip(x):
   """Flip the input x horizontally with 50% probability."""
   if np.random.rand(1)[0] > 0.5:
@@ -315,7 +320,7 @@ def _shear_x_impl(pil_img, level, img_shape):
   if random.random() > 0.5:
     level = -level
   return pil_img.transform(
-      (img_shape[0], img_shape[1]),
+      _width_height_from_img_shape(img_shape),
       Image.AFFINE,
       (1, level, 0, 0, 1, 0))
 
@@ -341,7 +346,7 @@ def _shear_y_impl(pil_img, level, img_shape):
   if random.random() > 0.5:
     level = -level
   return pil_img.transform(
-      (img_shape[0], img_shape[1]),
+      _width_height_from_img_shape(img_shape),
       Image.AFFINE,
       (1, 0, 0, level, 1, 0))
 
@@ -367,7 +372,7 @@ def _translate_x_impl(pil_img, level, img_shape):
   if random.random() > 0.5:
     level = -level
   return pil_img.transform(
-      (img_shape[0], img_shape[1]),
+      _width_height_from_img_shape(img_shape),
       Image.AFFINE,
       (1, 0, level, 0, 1, 0))
 
@@ -393,7 +398,7 @@ def _translate_y_impl(pil_img, level, img_shape):
   if random.random() > 0.5:
     level = -level
   return pil_img.transform(
-      (img_shape[0], img_shape[1]),
+      _width_height_from_img_shape(img_shape),
       Image.AFFINE,
       (1, 0, 0, 0, 1, level))