Miscs

correction.py

@@ -623,25 +623,6 @@ def process_single_task(task_tuple):
                             needs_correction.append(i)
                             break
 
-                        #
-                        # if ymin < yming-50 or ymax > ymaxg+50:
-                        #     print("Error : Gemini answered box2d too low/up", pid, label, group_name)
-                        #     if ymax < yming or ymin > ymaxg:
-                        #         print("Removing the box.")
-                        #         f["box_2d"] = None
-                        #         continue
-                        #     nymin = max(ymin, yming) * 1000 // total_height
-                        #     nymax = min(ymax, ymaxg) * 1000 // total_height
-                        #     f["box_2d"] = [nymin, xmin, nymax, xmax]
-
-                        # if f["box_2d"] and xmax / 1000 > width_r:
-                        #     print("Error : Gemini answered box2d too right", pid, label, group_name)
-                        #     if xmin / 1000 > width_r:
-                        #         print("Removing the box.")
-                        #         f["box_2d"] = None
-                        #         continue
-                        #     f["box_2d"][3] = int(width_r * 1000)
-
                 if needs_correction:
                     tprint(f"\tBox anomalies detected for Copie {pid} {group_name}. \n\tRequesting isolated correction from Gemini Flash...")
                     try:

gemini_for_labels.py

@@ -211,7 +211,7 @@ for path_str in args.input_paths:
 labels_txt = (INPUT_DIR / "labels").read_text()
 valid_labels_set = set(line.strip() for line in labels_txt.splitlines() if line.strip())
 names_path = (INPUT_DIR / "names")
-if !os.path.exists(names_path):
+if not os.path.exists(names_path):
     names_path = Path("names")
 names_txt = names_path.read_text()
 

plotting.py

@@ -9,7 +9,7 @@ from tkinter import messagebox
 from pathlib import Path
 from PIL import Image, ImageDraw, ImageFont, ImageTk
 
-print("o to open pdf, O original pdf, e to emacs part, click for coordinates")
+print("o to open pdf, O original pdf, e to emacs part, i to interro, click for coordinates")
 
 # --- Configuration & Globals ---
 padding = 60
@@ -54,7 +54,7 @@ def convert_list(l, group_id, json_schema):
         ll.append(ee)
     return ll
 
-def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages):
+def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages, last_label_index):
     im = Image.open(image_path)
     im.load()
     width, height = im.size
@@ -62,7 +62,6 @@ def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages):
     new_im.paste(im, (0, 0))
     draw = ImageDraw.Draw(new_im)
     bounding_boxes.sort(key=lambda b: (page_number(b["box_2d"], nb_pages), b["box_2d"][0]))
-    last_label_index = -1
 
     for bbox in bounding_boxes:
         raw_y_min = int(bbox["box_2d"][0] * height / 1000)
@@ -88,7 +87,7 @@ def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages):
                 draw.text((abs_x_min + 8, abs_y_min - 30), label, fill=color, font=font)
             else:
                 draw.text((abs_x_min + 8, abs_y_max + 6), label, fill=color, font=font)
-    return new_im
+    return (new_im, last_label_index)
 
 # --- Processing Logic (Worker Thread) ---
 
@@ -97,10 +96,15 @@ def worker_thread(base_dir, files_to_process, all_labels):
     Iterates through files, prepares VISUALS only, and puts metadata in queue.
     Does NOT write final JSON files anymore.
     """
+    previous_copie = None
+    last_label_index = None
     for img_path in files_to_process:
         json_path = base_dir / f"{img_path.stem}.json"
         copie_part = int(img_path.stem[-2:])
         copie = img_path.stem[:-3]
+        if copie != previous_copie:
+            last_label_index = -1
+            previous_copie = copie
         json_schema_path = base_dir / 'Cutleft' / f"{copie}_schema.json"
 
         try:
@@ -127,7 +131,8 @@ def worker_thread(base_dir, files_to_process, all_labels):
 
             try:
                 print(f"Buffering {img_path.name}...")
-                pil_image = prepare_image(str(img_path), bb_list, all_labels, nb_pages)
+                (pil_image, last_label_index) = \
+                    prepare_image(str(img_path), bb_list, all_labels, nb_pages, last_label_index)
 
                 metadata = {
                     "copie": copie,
@@ -169,6 +174,7 @@ class ImageViewer:
         self.root.bind('<Return>', self.on_enter)
         self.root.bind('e', self.on_edit)
         self.root.bind('o', self.on_open_pdf)
+        self.root.bind('i', self.on_open_interro)
         self.root.bind('O', self.on_open_ori_pdf)
         self.root.bind('<Escape>', lambda e: self.root.quit())
         self.label.bind('<Button-1>', self.on_click)
@@ -265,6 +271,12 @@ class ImageViewer:
             subprocess.Popen(['xdg-open', str(pdf_path.absolute())])
 
     def on_open_ori_pdf(self, event):
+        if self.is_viewing and self.current_json_path:
+            pdf_path = "/home/sebastien/Staging/Interro/" + base_dir + "pdf"
+            print(f"Opening {pdf_path}")
+            subprocess.Popen(['xdg-open', pdf_path])
+
+    def on_open_interro(self, event):
         if self.is_viewing and self.current_json_path:
             new_filename = self.current_json_path.stem.split('_')[0] + ".pdf"
             pdf_path = self.current_json_path.parent / "Copies Originales" / new_filename

post-correction.py

@@ -20,28 +20,28 @@ import ftfy
 import re
 import urllib.request
 
-# url = "https://raw.githubusercontent.com/hbenbel/French-Dictionary/master/dictionary/dictionary.txt"
-# french_words = urllib.request.urlopen(url).read().decode('utf-8').splitlines()
+with open('liste_francais.txt', 'r') as f:
+    french_words = f.read().splitlines()
 
 # 2. Pre-compute an O(1) lookup dictionary
 # We simulate the corruption by replacing accents with null bytes (\x00)
-# lookup_map = {}
-# for word in french_words:
-#     # Replace all French accents with \x00 to create the "broken" key
-#     broken_key = re.sub(r'[éèêëàâäîïôöùûüçœÉÈÊËÀÂÄÎÏÔÖÙÛÜÇŒ]', '\x00', word)
-#     if '\x00' in broken_key:
-#         lookup_map[broken_key] = word # e.g., "\x00cole" -> "école"
+lookup_map = {}
+for word in french_words:
+    # Replace all French accents with \x00 to create the "broken" key
+    broken_key = re.sub(r'[éèêëàâäîïôöùûüçœÉÈÊËÀÂÄÎÏÔÖÙÛÜÇŒ]', '\x00', word)
+    if '\x00' in broken_key:
+        lookup_map[broken_key] = word # e.g., "\x00cole" -> "école"
 
 # 3. Fast replace function
 def fast_fix(text):
     # Find words containing regular letters and null bytes
-    # def replacer(match):
-    #     broken_word = match.group(0)
-    #     # Return the fixed word from our map, or leave it if not found
-    #     # (Handles case-insensitivity by falling back to lowercase map)
-    #     return lookup_map.get(broken_word.lower(), broken_word)
+    def replacer(match):
+        broken_word = match.group(0)
+        # Return the fixed word from our map, or leave it if not found
+        # (Handles case-insensitivity by falling back to lowercase map)
+        return lookup_map.get(broken_word.lower(), broken_word)
 
-    # return re.sub(r'[a-zA-Z\x00]+', replacer, text)
+    return re.sub(r'[a-zA-Z\x00]+', replacer, text)
     return text