#!/usr/bin/env python3 import sys sys.path.insert(0, '/Users/guillemhernandezsola/code/manga-translator') import cv2 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) # Filter filtered = [] for bbox, text, conf in raw: t = mt.normalize_text(text) if conf < 0.12 or len(t) < 1 or mt.is_noise_text(t) or mt.is_sound_effect(t) or mt.is_title_text(t): continue filtered.append((bbox, t, conf)) # Get the indices we're interested in (left and right bubbles) left_indices = [41, 42, 43, 44, 45, 46] # LET, GO, OFF, ME, AL-, REA- right_indices = [47, 48, 49, 50, 51, 52, 53, 54] # DON'T, WORRY!, HARUKO, ... print("=== CHECKING GROUPING CONDITIONS ===\n") # Check if they would be united in group_tokens boxes_left = [mt.quad_bbox(filtered[i][0]) for i in left_indices] boxes_right = [mt.quad_bbox(filtered[i][0]) for i in right_indices] # Check overlap_or_near print("Checking overlap_or_near with gap=18:") for li, bi in enumerate(left_indices): for ri, bj in enumerate(right_indices): b_left = boxes_left[li] b_right = boxes_right[ri] gap_x = max(0, max(b_left[0], b_right[0]) - min(b_left[2], b_right[2])) gap_y = max(0, max(b_left[1], b_right[1]) - min(b_left[3], b_right[3])) overlaps = gap_x <= 18 and gap_y <= 18 if overlaps: print(f" {bi} and {bj} overlap/near: gap_x={gap_x}, gap_y={gap_y}") # Check distance check hs = [max(1.0, b[3] - b[1]) for b in [*boxes_left, *boxes_right]] med_h = float(np.median(hs)) if hs else 12.0 dist_thresh = max(20.0, med_h * 2.2) print(f"\nMedian height: {med_h}") print(f"Distance threshold: {dist_thresh}") print("\nChecking distance check:") for li, bi in enumerate(left_indices[:1]): # Just check first from each for ri, bj in enumerate(right_indices[:1]): b_left = boxes_left[li] b_right = boxes_right[ri] cx_left = (b_left[0] + b_left[2]) / 2.0 cy_left = (b_left[1] + b_left[3]) / 2.0 cx_right = (b_right[0] + b_right[2]) / 2.0 cy_right = (b_right[1] + b_right[3]) / 2.0 d = ((cx_left - cx_right) ** 2 + (cy_left - cy_right) ** 2) ** 0.5 within_dist = d <= dist_thresh within_y = abs(cy_left - cy_right) <= med_h * 3.0 print(f" {bi} to {bj}: distance={d:.1f}, within_dist={within_dist}, within_y_tol={within_y}")