Genarrative/tools/analyze_puzzle_clear_output.py

"""
分析拼消消测试输出，检测卡片透明区域并溯源问题成因。

用法:
  python tools/analyze_puzzle_clear_output.py
  python tools/analyze_puzzle_clear_output.py --dir path/to/output
  python tools/analyze_puzzle_clear_output.py --detail  # 详细逐卡输出
"""

import os
import sys
import argparse
from collections import defaultdict
from PIL import Image


def scan_transparent_pixels(img_path):
    """扫描图片，返回 (总像素数, 透明像素数, 边缘透明列比例)"""
    img = Image.open(img_path).convert("RGBA")
    w, h = img.size
    pixels = img.load()
    total = w * h
    transparent = 0
    edge_cols_with_transparent = 0
    edge_rows_with_transparent = 0

    # 统计透明像素
    for y in range(h):
        for x in range(w):
            if pixels[x, y][3] < 128:
                transparent += 1

    # 检测四边是否有透明像素（边缘列/行透明占比 > 10%）
    for x in range(w):
        col_transparent = sum(1 for y in range(h) if pixels[x, y][3] < 128)
        if col_transparent > h * 0.1:
            edge_cols_with_transparent += 1

    for y in range(h):
        row_transparent = sum(1 for x in range(w) if pixels[x, y][3] < 128)
        if row_transparent > w * 0.1:
            edge_rows_with_transparent += 1

    ratio = transparent / total * 100 if total > 0 else 0
    has_edge = edge_cols_with_transparent > 0 or edge_rows_with_transparent > 0
    return total, transparent, ratio, has_edge, edge_cols_with_transparent, edge_rows_with_transparent


def analyze_sheet_cleaned(sheet_path):
    """分析去背后的 sheet 图，检查各 group 区域的透明情况"""
    img = Image.open(sheet_path).convert("RGBA")
    w, h = img.size
    pixels = img.load()

    # 统计整体透明像素
    total = w * h
    transparent = sum(1 for y in range(h) for x in range(w) if pixels[x, y][3] < 128)
    return total, transparent, transparent / total * 100 if total > 0 else 0


def main():
    parser = argparse.ArgumentParser(description="分析拼消消测试输出")
    parser.add_argument("--dir", default="", help="输出目录路径")
    parser.add_argument("--detail", action="store_true", help="详细逐卡输出")
    args = parser.parse_args()

    # 自动查找输出目录
    if args.dir:
        base = args.dir
    else:
        script_dir = os.path.dirname(os.path.abspath(__file__))
        repo_root = os.path.dirname(script_dir)
        candidates = [
            os.path.join(repo_root, "server-rs", "crates", "api-server", "target", "test-output", "puzzle-clear-real"),
            os.path.join(repo_root, "server-rs", "target", "test-output", "puzzle-clear-real"),
        ]
        base = None
        for c in candidates:
            if os.path.isdir(c):
                base = c
                break
        if not base:
            print("未找到测试输出目录。请用 --dir 指定路径。")
            sys.exit(1)

    sheets_dir = os.path.join(base, "sheets")
    cards_dir = os.path.join(base, "cards")

    if not os.path.isdir(cards_dir):
        print(f"cards 目录不存在: {cards_dir}")
        sys.exit(1)

    # ==================== 阶段 1: 分析卡片 ====================
    print("=" * 70)
    print("阶段 1: 卡片透明像素分析")
    print("=" * 70)

    card_results = []  # (sheet, card_name, total, transparent, ratio, has_edge, edge_cols, edge_rows)
    problem_cards = []

    for sheet_name in sorted(os.listdir(cards_dir)):
        sheet_dir = os.path.join(cards_dir, sheet_name)
        if not os.path.isdir(sheet_dir):
            continue
        for card_name in sorted(os.listdir(sheet_dir)):
            if not card_name.endswith(".png"):
                continue
            card_path = os.path.join(sheet_dir, card_name)
            total, trans, ratio, has_edge, ec, er = scan_transparent_pixels(card_path)
            card_results.append((sheet_name, card_name, total, trans, ratio, has_edge, ec, er))
            if ratio > 5 or has_edge:
                problem_cards.append((sheet_name, card_name, total, trans, ratio, has_edge, ec, er))

    # 按 sheet 汇总
    by_sheet = defaultdict(list)
    for r in card_results:
        by_sheet[r[0]].append(r)

    print(f"\n总卡片数: {len(card_results)}")
    print(f"问题卡片数 (透明>5% 或 有边缘透明): {len(problem_cards)}")
    print()

    for sheet_name in sorted(by_sheet.keys()):
        cards = by_sheet[sheet_name]
        problem_count = sum(1 for r in cards if r[4] > 5 or r[5])
        print(f"  {sheet_name}: {len(cards)} cards, {problem_count} problems")

    if problem_cards:
        print(f"\n--- 问题卡片详情 ---")
        problem_cards.sort(key=lambda r: -r[4])  # sort by ratio desc
        for sheet, name, total, trans, ratio, has_edge, ec, er in problem_cards:
            group_id = name.split("-part-")[0]
            edge_info = f", 边缘透明列={ec} 行={er}" if has_edge else ""
            print(f"  {sheet}/{name}  group={group_id}  transparent={ratio:.1f}% ({trans}/{total}){edge_info}")

    # ==================== 阶段 2: 溯源分析 ====================
    print()
    print("=" * 70)
    print("阶段 2: 溯源 — 对比原始 sheet 与去背后 sheet")
    print("=" * 70)

    if os.path.isdir(sheets_dir):
        for fname in sorted(os.listdir(sheets_dir)):
            if not fname.endswith(".png"):
                continue
            sheet_path = os.path.join(sheets_dir, fname)
            total, trans, ratio = analyze_sheet_cleaned(sheet_path)
            is_cleaned = "-cleaned" in fname
            label = "去背后" if is_cleaned else "原始"
            print(f"  {fname} ({label}): {trans}/{total} 透明像素 ({ratio:.1f}%)")

    # ==================== 阶段 3: 问题溯源推理 ====================
    print()
    print("=" * 70)
    print("阶段 3: 问题成因分析")
    print("=" * 70)

    if not problem_cards:
        print("  无问题卡片，管线正常。")
        return

    # 分析问题卡片的 group 分布
    problem_groups = defaultdict(list)
    for r in problem_cards:
        group_id = r[1].split("-part-")[0]
        problem_groups[group_id].append(r)

    print(f"\n  涉及 {len(problem_groups)} 个 group:")
    for group_id in sorted(problem_groups.keys()):
        cards = problem_groups[group_id]
        avg_ratio = sum(r[4] for r in cards) / len(cards)
        edge_count = sum(1 for r in cards if r[5])
        print(f"    {group_id}: {len(cards)} cells, avg透明={avg_ratio:.1f}%, {edge_count} cells有边缘透明")

    # 检查原始 sheet 和去背后 sheet 的差异
    if os.path.isdir(sheets_dir):
        cleaned_files = [f for f in os.listdir(sheets_dir) if "cleaned" in f]
        raw_files = [f for f in os.listdir(sheets_dir) if "cleaned" not in f and f.endswith(".png")]
        if cleaned_files and raw_files:
            for raw_f in sorted(raw_files):
                raw_p = os.path.join(sheets_dir, raw_f)
                cleaned_f = raw_f.replace(".png", "-cleaned.png")
                cleaned_p = os.path.join(sheets_dir, cleaned_f)
                if not os.path.exists(cleaned_p):
                    continue
                _, raw_trans, raw_ratio = analyze_sheet_cleaned(raw_p)
                _, cleaned_trans, cleaned_ratio = analyze_sheet_cleaned(cleaned_p)
                print(f"\n    {raw_f}:")
                print(f"      原始透明: {raw_ratio:.1f}%")
                print(f"      去后透明: {cleaned_ratio:.1f}%")
                delta = cleaned_ratio - raw_ratio
                if delta > 1:
                    print(f"      ** 去背增加了 {delta:.1f}% 透明像素 — 可能误删了主体内容")

    print()
    print("  可能原因:")
    print("    1. AI 未将内容画满整个 group 区域（内容在组内偏移）")
    print("    2. 洋红去背误将主题内近似洋红的像素也变透明")
    print("    3. find_non_transparent_bounds 扫描范围包括了相邻 group 的透明间隙")
    print("    4. group resize 后未完全覆盖目标尺寸（内容比例与目标比例不匹配）")


if __name__ == "__main__":
    main()