chore(docs): document ScoreExtractor tiling and refactor debug scripts (#563)

scripts/debug/verify_monotonic.py

@@ -0,0 +1,41 @@
+import sys
+sys.path.append(r"C:\Users\Certes\Desktop\guitar_score")
+import cv2
+import easyocr
+import numpy as np
+import os
+from youtube_tab_to_pdf import extract_frames, extract_unique_scroll
+
+video_file = r"C:\Users\Certes\Desktop\guitar_score\output\サカナクション／新宝島(エレキギターTAB) 難易度★★★ sakanaction shintakarajima.mp4"
+print("Extracting frames...")
+frames = extract_frames(video_file, fps=2)
+
+print("Running pipeline extraction...")
+unique = extract_unique_scroll(frames, threshold=0.95)
+
+print("Initializing OCR...")
+reader = easyocr.Reader(['en'])
+
+print(f"Generated {len(unique)} chunks.")
+detect_log = []
+
+for i, page in enumerate(unique):
+    # Image is A4 width
+    # We want to OCR the top 150 pixels of the whole chunk to find measure numbers
+    h, w = page.shape[:2]
+    top_area = page[:min(200, h), :]
+    
+    results = reader.readtext(top_area)
+    # filter for numbers
+    nums = []
+    for (bbox, text, prob) in results:
+        t = ''.join(filter(str.isdigit, text))
+        if t:
+            nums.append(int(t))
+            
+    print(f"Page {i} measure numbers detected: {nums}")
+    detect_log.append(nums)
+    
+    cv2.imwrite(f"output/verify_chunk_{i}.jpg", page)
+    if i > 5:
+        break