Fix version 0.2.0-dev3 and later

In this MR, version 0.2.0-dev3 (and later) is fixed:

• We use DPIAdjusting transform as it is a key component when training on images of very different sizes
• We compute the real mean and std values on the train set (we do not use default values from ImageNet = revert a5f7b5ab)
• The padding is on the right and bottom of the image (= revert part of f57bc667 for now)
• Two parameters files are created during training: one for training details (used to reproduce/continue the experiment), one for inference (that can be directly used by Arkindex DAN worker)

dan/utils.py

@@ -26,11 +26,11 @@ def pad_sequences_1D(data, padding_value):
     return padded_data
 
 
-def pad_images(data):
+def pad_images(images):
     """
-    Pad the images so that they are in the middle of the large padded image (tb-lr mode).
-    :param data: List of numpy arrays.
-    :return padded_data: A tensor containing all the padded images.
+    Pad the images so that they are at the top left of the large padded image.
+    :param images: List of images as torch tensors.
+    :return padded_images: A tensor containing all the padded images.
     """
     longest_x = max([x.shape[1] for x in data])
     longest_y = max([x.shape[2] for x in data])
@@ -46,7 +46,7 @@ def pad_images(data):
             top : padded_data.shape[2] - bottom,
             left : padded_data.shape[3] - right,
         ] = image
-    return padded_data
+    return padded_images
 
 
 def read_image(path):