Everything

2026-04-22 16:27:56 +02:00
parent b730037a06
commit d77db83cfe
13 changed files with 2 additions and 4259 deletions
--- a/older-code/debug_split_phase.py
+++ b/older-code/debug_split_phase.py
@@ -0,0 +1,107 @@
+#!/usr/bin/env python3
+# Debug script to see what bubbles are produced after splitting
+
+import sys
+sys.path.insert(0, '/Users/guillemhernandezsola/code/manga-translator')
+
+import cv2
+import json
+import numpy as np
+import importlib.util
+
+spec = importlib.util.spec_from_file_location("manga_translator", "/Users/guillemhernandezsola/code/manga-translator/manga-translator.py")
+mt = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(mt)
+
+image_path = '004.png'
+detector = mt.MacVisionDetector(source_lang='en')
+raw = detector.read(image_path)
+image = cv2.imread(image_path)
+
+# Full filtering as pipeline does
+filtered = []
+skipped = 0
+ih, iw = image.shape[:2]
+
+for bbox, text, conf in raw:
+    t = mt.normalize_text(text)
+    qb = mt.quad_bbox(bbox)
+    
+    if conf < 0.12:
+        skipped += 1
+        continue
+    if len(t) < 1:
+        skipped += 1
+        continue
+    if mt.is_noise_text(t):
+        skipped += 1
+        continue
+    if mt.is_sound_effect(t):
+        skipped += 1
+        continue
+    if mt.is_title_text(t):
+        skipped += 1
+        continue
+    if qb[1] < int(ih * mt.TOP_BAND_RATIO):
+        if conf < 0.70 and len(t) >= 5:
+            skipped += 1
+            continue
+
+    filtered.append((bbox, t, conf))
+
+resolved_gap = mt.auto_gap(image_path)
+bubbles, bubble_boxes, bubble_quads, bubble_indices = mt.group_tokens(
+    filtered, image.shape, gap_px=resolved_gap, bbox_padding=3
+)
+
+print("=== AFTER GROUPING ===")
+print(f"Bubbles dict keys: {sorted(bubbles.keys())}")
+for bid in [7, 8]:
+    if bid in bubbles:
+        print(f"\nBubble {bid}:")
+        print(f"  Box: {bubble_boxes[bid]}")
+        print(f"  Indices ({len(bubble_indices[bid])}): {bubble_indices[bid]}")
+        print(f"  Quads ({len(bubble_quads[bid])})")
+
+# Now simulate the split logic
+new_bubbles, new_bubble_boxes, new_bubble_quads, new_bubble_indices = {}, {}, {}, {}
+next_bid = max(bubbles.keys()) + 1 if bubbles else 1
+splits_performed = []
+
+for bid in list(bubbles.keys()):
+    box = bubble_boxes[bid]
+    bubble_split = None
+    
+    # Try split
+    split_result = mt.split_panel_box(image, box, bubble_quads=bubble_quads[bid])
+    if split_result:
+        box_left, box_right, split_x = split_result
+        # ... split logic ...
+        bubble_split = "panel_split"
+    
+    if bubble_split is None:
+        col_split = mt.split_bubble_if_multiple_columns(bubble_indices[bid], filtered, bid=bid)
+        if col_split:
+            bubble_split = "column_split"
+    
+    if bubble_split:
+        splits_performed.append(f"Bubble {bid}: {bubble_split}")
+        # Don't actually split here, just mark it
+    else:
+        # No split
+        new_bubbles[bid] = bubbles[bid]
+        new_bubble_boxes[bid] = bubble_boxes[bid]
+        new_bubble_quads[bid] = bubble_quads[bid]
+        new_bubble_indices[bid] = bubble_indices[bid]
+
+print("\n=== AFTER SPLIT LOGIC ===")
+print(f"Splits detected: {len(splits_performed)}")
+for s in splits_performed:
+    print(f"  {s}")
+    
+print(f"\nBubbles dict keys: {sorted(new_bubbles.keys())}")
+for bid in [7, 8]:
+    if bid in new_bubbles:
+        print(f"\nBubble {bid}:")
+        print(f"  Box: {new_bubble_boxes[bid]}")
+        print(f"  Indices ({len(new_bubble_indices[bid])}): {new_bubble_indices[bid][:3]}...")