docs: add information for the argument separate (#732)

Author: sageyou

mindcv/data/loader.py

@@ -58,7 +58,11 @@ def create_loader(
             (default=None).
         python_multiprocessing (bool, optional): Parallelize Python operations with multiple worker processes. This
             option could be beneficial if the Python operation is computational heavy (default=False).
-        separate(bool, optional): separate the image origin and the image been transformed
+        separate(bool, optional): separate the image clean and the image been transformed.
+            If separate==True, that means the dataset returned has 3 parts:
+            * the first part called image "clean", which means the image without auto_augment (e.g., auto-aug)
+            * the second and third parts called image transformed, hence, with the auto_augment transform.
+            Refer to ".transforms_factory.create_transforms" for more information.
 
     Note:
         1. cutmix is now experimental (which means performance gain is not guarantee)

mindcv/data/transforms_factory.py

@@ -196,7 +196,11 @@ def create_transforms(
         image_resize (int): the image size after resize for adapting to network. Default: 224.
         is_training (bool): if True, augmentation will be applied if support. Default: False.
         auto_augment(str)：augmentation strategies, such as "augmix", "autoaug" etc.
-        separate: separate the image origin and the image been transformed.
+        separate: separate the image clean and the image been transformed. If separate==True, the transformers are
+            returned as a tuple of 3 separate transforms for use in a mixing dataset that  passes:
+            * all data through the primary transform, called "clean" data
+            * a portion of the data through the secondary transform (e.g., auto-aug)
+            * normalized and converts the branches above with the third, transform
         **kwargs: additional args parsed to `transforms_imagenet_train` and `transforms_imagenet_eval`
 
     Returns: