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

scripts/debug/test_number_band.py

@@ -0,0 +1,33 @@
+import cv2
+import numpy as np
+from youtube_tab_to_pdf import extract_unique_scroll
+
+# We will read fast_test_pano.jpg
+img = cv2.imread('fast_test_pano.jpg', cv2.IMREAD_GRAYSCALE)
+
+# We want to find staff lines and number band
+_, bin_inv = cv2.threshold(img, 200, 255, cv2.THRESH_BINARY_INV)
+row_sums = np.sum(bin_inv, axis=1) / 255.0
+staff_rows = np.where(row_sums > img.shape[1] * 0.4)[0]
+
+if len(staff_rows) >= 6:
+    staff_y_top = staff_rows[0]
+else:
+    staff_y_top = int(img.shape[0] * 0.3)
+
+# Number band
+band_y_top = max(0, staff_y_top - 25)
+band_y_bottom = staff_y_top
+
+band = img[band_y_top:band_y_bottom, :]
+
+# Save it to see if it correctly contains the numbers
+cv2.imwrite('debug_band.png', band)
+print(f"Band shape: {band.shape}")
+
+# Let's see if we can extract number boxes!
+band_inv = cv2.bitwise_not(band)
+col_sums = np.sum(band_inv, axis=0) / 255.0
+number_xs = np.where(col_sums > 5)[0] # at least 5 pixels of ink vertically
+
+print(f"Pixels with numbers: {len(number_xs)}")