Rename images with line id

Two main changes:

• Images are renamed {page_id}_{line_number}_{line_id} (previously {page_id}_{line_number}) as it is easier to visualize on Arkindex
• Line number (e.g line_032) should come from the element's name and not from enumerate to avoid offset on pages containing lines without transcription

Note: I'm assuming that there are at most 999 lines/image (line_number has 3 digits)

kaldi_data_generator/main.py

@@ -165,10 +165,14 @@ class HTRDataGenerator:
     def get_accepted_zones(self, page_id: str):
         try:
             accepted_zones = []
+            print(self.api_client.cached_paginate(
+                "ListElementChildren", id=page_id, with_classes=True
+            ))
             for elt in self.api_client.cached_paginate(
                 "ListElementChildren", id=page_id, with_classes=True
             ):
                 elem_classes = [c for c in elt["classes"] if c["state"] != "rejected"]
+                print(elt, elem_classes)
 
                 should_accept = True
                 if self.should_filter_by_class:
@@ -307,6 +311,7 @@ class HTRDataGenerator:
 
                 trans_data = TranscriptionData(
                     element_id=res["element"]["id"],
+                    element_name=res["element"]["name"],
                     polygon=polygon,
                     text=text,
                     trans_id=res["id"],
@@ -380,18 +385,25 @@ class HTRDataGenerator:
         }
 
     def _save_line_image(
-        self, page_id, i, line_img, manifest_fp=None, trans: TranscriptionData = None
+        self, page_id, line_img, manifest_fp=None, trans: TranscriptionData = None
     ):
+        # Get line id
+        line_id = trans.element_id
+
+        # Get line number from its name
+        line_number = trans.element_name.split('_')[-1]
+
         if self.should_rotate:
             if trans.rotation_class:
                 rotate_angle = ROTATION_CLASSES_TO_ANGLES[trans.rotation_class]
                 line_img = rotate_and_trim(line_img, rotate_angle, WHITE)
         if self.format == "kraken":
-
-            save_img(f"{self.out_line_dir}/{page_id}_{i}.png", line_img)
-            manifest_fp.write(f"{page_id}_{i}.png\n")
+            # Save image using the template {page_id}_{line_number}_{line_id}
+            # TODO: check if (0>3) is enough (pad line_number to 3 digits)
+            save_img(f"{self.out_line_dir}/{page_id}_{line_number:0>3}_{line_id}.png".format(), line_img)
+            manifest_fp.write(f"{page_id}_{line_number:0>3}_{line_id}.png\n")
         else:
-            save_img(f"{self.out_line_img_dir}/{page_id}_{i}.jpg", line_img)
+            save_img(f"{self.out_line_img_dir}/{page_id}_{line_number:0>3}_{line_id}.jpg", line_img)
 
     def extract_lines(self, page_id: str, image_data: dict):
         if self.should_filter_by_class or self.should_filter_by_style:
@@ -441,7 +453,7 @@ class HTRDataGenerator:
             # not needed for kaldi
             manifest_fp = None
 
-        for i, trans in enumerate(sorted_lines):
+        for trans in sorted_lines:
             extracted_img = extract(
                 img=img,
                 polygon=trans.polygon,
@@ -452,7 +464,8 @@ class HTRDataGenerator:
                 grayscale=self.grayscale,
             )
 
-            self._save_line_image(page_id, i, extracted_img, manifest_fp, trans)
+            # don't enumerate, read the line number from the elements's name (e.g. line_xx) so that it matches with Arkindex 
+            self._save_line_image(page_id, extracted_img, manifest_fp, trans)
 
         if self.format == "kraken":
             manifest_fp.close()