首次提交

tools/rerun_srt.py

@@ -0,0 +1,99 @@
+# -*- coding: utf-8 -*-
+"""
+从已有 OCR 数据重新生成 SRT，不重新调用 OCR API。
+用法: python rerun_srt.py
+"""
+import sys, os, json
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+from difflib import SequenceMatcher
+from tool.log.log_manager import UnifiedLogger
+from tool.getsrt import (
+    analyze_subtitle_blocks, analyze_h,
+    extract_highest_subtitle, check_subtitle_quality_enhanced,
+    evaluate_text_quality_rules, generate_srt_from_results
+)
+
+COMPOSITE_DIR = sys.argv[1] if len(sys.argv) > 1 else "./composite/BV1aL4y1p7Bi_p11"
+JSONL_PATH    = os.path.join(COMPOSITE_DIR, "ocr_result.jsonl")
+META_PATH     = os.path.join(COMPOSITE_DIR, "metadata.json")
+OUTPUT_SRT    = os.path.join(COMPOSITE_DIR, "BV1aL4y1p7Bi_p11.new.srt")
+INTERVAL      = 0.33
+
+logger = UnifiedLogger()
+logger.reset()
+logger.init_logging("rerun_log.json")
+logger.start_program("rerun")
+
+# 读取已有 OCR 数据，将旧路径修正为当前实际路径
+all_pic_ocr_pair = []
+with open(JSONL_PATH, 'r', encoding='utf-8') as f:
+    lines = f.readlines()
+
+i = 0
+while i < len(lines):
+    if i + 1 < len(lines) and lines[i].startswith("pic_path: "):
+        raw_path = lines[i].strip()[len("pic_path: "):]
+        filename = os.path.basename(raw_path)
+        actual_path = os.path.join(COMPOSITE_DIR, filename)
+        try:
+            ocr_json = json.loads(lines[i + 1])
+            all_pic_ocr_pair.append([actual_path, ocr_json])
+            i += 2
+            continue
+        except json.JSONDecodeError:
+            pass
+    i += 1
+
+print(f"读取 {len(all_pic_ocr_pair)} 帧的OCR数据")
+
+# 分析字幕网格位置
+best_grids = analyze_subtitle_blocks(all_pic_ocr_pair, META_PATH, logger)
+h = analyze_h(all_pic_ocr_pair, META_PATH)
+
+if not best_grids.get("composite"):
+    print("ERROR: 未找到合适的字幕区域")
+    sys.exit(1)
+
+# 逐帧提取字幕
+results = []
+DEBUG_KEYWORD = "曹小月从角落里拿出雨伞"
+for idx in range(len(all_pic_ocr_pair)):
+    print(f"\r处理 {idx + 1}/{len(all_pic_ocr_pair)}", end="", flush=True)
+    text, conf, flag, sim = extract_highest_subtitle(
+        idx, all_pic_ocr_pair, best_grids, META_PATH, h, logger
+    )
+    if text and not check_subtitle_quality_enhanced(text) and conf <= 90:
+        text = ""
+    # 打印包含关键词的帧的原始OCR和提取结果
+    if DEBUG_KEYWORD in (text or ""):
+        pic_path, ocr_json = all_pic_ocr_pair[idx]
+        print(f"\n[DEBUG 帧{idx}] 提取: {repr(text)}")
+        for td in ocr_json['TextDetections']:
+            print(f"  box: {repr(td['DetectedText'])}  conf={td['Confidence']}  watermark={td.get('is_watermark','?')}")
+    results.append((idx, text or ""))
+
+non_empty = sum(1 for _, t in results if t)
+print(f"\n共 {non_empty}/{len(results)} 帧有字幕")
+
+# 生成 SRT
+generate_srt_from_results(results, OUTPUT_SRT, interval_seconds=INTERVAL)
+print(f"SRT 已输出: {OUTPUT_SRT}")
+
+# 展示问题字幕附近的结果
+with open(OUTPUT_SRT, 'r', encoding='utf-8') as f:
+    srt_content = f.readlines()
+
+print("\n=== 修复验证（'曹小月从角落里拿出雨伞' 附近） ===")
+found = False
+for j, line in enumerate(srt_content):
+    if "曹小月从角落里拿出雨伞" in line:
+        start = max(0, j - 3)
+        end   = min(len(srt_content), j + 10)
+        print("".join(srt_content[start:end]))
+        print("---")
+        found = True
+        break
+
+if not found:
+    print("未在输出 SRT 中找到目标字幕，请检查提取结果。")