jira:NYJ-1540 desc: add video_event_for_contents func

lizhengwei
Commit 784c8effb9211d075ce6ea641718dcccb11ae660 784c8eff 1 parent 1396dd74
Showing 2 changed files with 96 additions and 53 deletions
src/football_replay_match/core/llm_image.py
src/football_replay_match/util/football_replay_video_event_by_llm.py
--- a/src/football_replay_match/core/llm_image.py
View file @784c8ef
+++ b/src/football_replay_match/core/llm_image.py
View file @784c8ef
@@ -122,7 +122,7 @@ class Video2Frame:
         if not cap.isOpened():
             raise ValueError(f"无法打开视频片段: {clip_path}")
-        total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
+        # total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
         video_fps = cap.get(cv2.CAP_PROP_FPS)
         if video_fps <= 0:
             video_fps = 25.0
@@ -132,10 +132,9 @@ class Video2Frame:
         if frame_interval < 1:
             frame_interval = 1
-        frame_paths = []
+        frames = []
         saved_count = 0
         frame_count = 0
-
         while True:
             ret, frame = cap.read()
             if not ret:
@@ -145,27 +144,26 @@ class Video2Frame:
                 if roi is not None:
                     x, y, w, h = roi
                     frame = frame[y:y + h, x:x + w]
-
-                if max_px_area is not None:
-                    h, w = frame.shape[:2]
-                    area = h * w
-                    if area > max_px_area:
-                        scale = (max_px_area / area) ** 0.5
-                        new_w = int(w * scale)
-                        new_h = int(h * scale)
-                        frame = cv2.resize(
-                            frame, (new_w, new_h), interpolation=cv2.INTER_AREA
-                        )
-
+                frame = self.reisize_frame(frame, max_px_area)
+                frames.append(frame)
                 frame_path = frame_output_dir / f"frame_{saved_count:06d}.jpg"
                 cv2.imwrite(str(frame_path), frame)
-                frame_paths.append(str(frame_path))
                 saved_count += 1
-
             frame_count += 1
-
         cap.release()
-        return frame_paths
+        return frames
+
+    def reisize_frame(self, frame, max_px_area):
+        if max_px_area is None:
+            return frame
+        h, w = frame.shape[:2]
+        area = h * w
+        if area > max_px_area:
+            scale = (max_px_area / area) ** 0.5
+            new_w = int(w * scale)
+            new_h = int(h * scale)
+            frame = cv2.resize(frame, (new_w, new_h), interpolation=cv2.INTER_AREA)
+        return frame
     @staticmethod
     def _find_frame_idx(utc_list, target_utc, find_last_not_greater=True):
@@ -241,28 +239,34 @@ def frames2content(frames):
     将帧列表（图片路径列表）转为 LLM content 列表。
     格式与 llm_video_content.py 中的 contents 返回的 content 一致。
     """
-    content = []
+    contents = []
     video_prompt = (
         f"以下是从视频中按时间顺序提取的 {len(frames)} 帧画面，"
         f"请将它们视为一个连续的视频进行分析。"
     )
-    content.append({"type": "text", "text": video_prompt})
+    contents.append({"type": "text", "text": video_prompt})
-    for frame_path in frames:
-        with open(frame_path, "rb") as f:
-            img_bytes = f.read()
-        b64_str = base64.b64encode(img_bytes).decode("utf-8")
-        content.append({
+    for frame in frames:
+        _, buffer = cv2.imencode('.jpg', frame)
+        b64_str = base64.b64encode(buffer).decode('utf-8')
+
+        # with open(frame_path, "rb") as f:
+        #     img_bytes = f.read()
+        # b64_str = base64.b64encode(img_bytes).decode("utf-8")
+
+        contents.append({
             "type": "image_url",
             "image_url": {
                 "url": f"data:image/jpeg;base64,{b64_str}"
             }
         })
-    return content
+    return contents
 if __name__ == "__main__":
+    # pass
+
     import argparse
     import sys
     import tempfile
@@ -275,7 +279,7 @@ if __name__ == "__main__":
     )
     parser.add_argument(
         "--cache-dir",
-        default=None,
+        default=r"C:\Users\lzw\AppData\Local\Temp\video2frame_test_mn26tcy_my",
         help="缓存目录（默认自动创建临时目录）",
     )
     parser.add_argument(
@@ -293,7 +297,7 @@ if __name__ == "__main__":
     parser.add_argument(
         "--fps",
         type=float,
-        default=1.0,
+        default=2.0,
         help="抽帧帧率（默认 1.0）",
     )
     args = parser.parse_args()
@@ -314,11 +318,13 @@ if __name__ == "__main__":
             start_utc=args.start,
             end_utc=args.end,
             fps=args.fps,
+            roi=None,
+            max_px_area=200_000,
         )
         print(f"  成功提取 {len(frames)} 帧")
         if frames:
-            print(f"  首帧: {frames[0]}")
-            print(f"  末帧: {frames[-1]}")
+            print(f"  首帧: {frames[0].shape}")
+            print(f"  末帧: {frames[-1].shape}")
     except Exception as e:
         print(f"  失败: {e}")
         import traceback
@@ -340,27 +346,46 @@ if __name__ == "__main__":
     if len(content) > 3:
         print(f"  ... 还有 {len(content) - 3} 个元素")
-    # TEST 3: ROI + max_px_area
-    print("\n[TEST 3] to_frames with roi + max_px_area")
-    try:
-        end_crop = min(args.start + 5, args.end)
-        frames_cropped = v2f.to_frames(
-            url=args.url,
-            start_utc=args.start,
-            end_utc=end_crop,
-            fps=1.0,
-            roi=None,
-            max_px_area=200_000,
-        )
-        print(f"  成功提取 {len(frames_cropped)} 帧（带 ROI 裁剪和面积缩放）")
-    except Exception as e:
-        print(f"  跳过/失败: {e}")
+    print("\n[TEST 3] FootballReplayVideoEvent 进球识别（可选）")
-    # TEST 4: SEI UTC 模式说明（仅演示，需真实含 SEI 的流才能运行）
-    print("\n[TEST 4] SEI UTC 绝对时间戳模式（示例代码，需替换为含 SEI 的流）")
-    print(
-        "  # 示例：假设视频第一帧 UTC 为 1715600000，提取 10 秒片段\n"
-        "  # frames = v2f.to_frames(url, start_utc=1715600000, end_utc=1715600010, fps=1)"
+    try:
+        from ..util.football_replay_video_event_by_llm import FootballReplayVideoEvent
+    except ImportError:
+        import sys
+        from pathlib import Path
+        sys.path.insert(0, str(Path(__file__).parent.parent))
+        from util.football_replay_video_event_by_llm import FootballReplayVideoEvent
+
+    fbrv = FootballReplayVideoEvent(
+        base_url="http://192.168.1.59:11434",
+        model="qwen3.6:35b-a3b-q8_0",
+        temperature=0.7,
     )
-
-    print("\n[INFO] 所有测试完成")
+    event_json = fbrv.video_event_for_contents(content, None)
+    print(f"  识别结果: {event_json}")
+
+
+    # # TEST 3: ROI + max_px_area
+    # print("\n[TEST 3] to_frames with roi + max_px_area")
+    # try:
+    #     end_crop = min(args.start + 5, args.end)
+    #     frames_cropped = v2f.to_frames(
+    #         url=args.url,
+    #         start_utc=args.start,
+    #         end_utc=end_crop,
+    #         fps=1.0,
+    #         roi=None,
+    #         max_px_area=200_000,
+    #     )
+    #     print(f"  成功提取 {len(frames_cropped)} 帧（带 ROI 裁剪和面积缩放）")
+    # except Exception as e:
+    #     print(f"  跳过/失败: {e}")
+    #
+    # # TEST 4: SEI UTC 模式说明（仅演示，需真实含 SEI 的流才能运行）
+    # print("\n[TEST 4] SEI UTC 绝对时间戳模式（示例代码，需替换为含 SEI 的流）")
+    # print(
+    #     "  # 示例：假设视频第一帧 UTC 为 1715600000，提取 10 秒片段\n"
+    #     "  # frames = v2f.to_frames(url, start_utc=1715600000, end_utc=1715600010, fps=1)"
+    # )
+    #
+    # print("\n[INFO] 所有测试完成")
--- a/src/football_replay_match/util/football_replay_video_event_by_llm.py
View file @784c8ef
+++ b/src/football_replay_match/util/football_replay_video_event_by_llm.py
View file @784c8ef
@@ -96,3 +96,21 @@ class FootballReplayVideoEvent:
             with open(cache_path, 'w', encoding='utf-8') as f:
                 f.write(result)
         return result_json
+
+    def video_event_for_contents(self, contents: list, asr_text: str = '无', cache_path=None):
+        if cache_path is not None and os.path.exists(cache_path):
+            with open(cache_path, 'r', encoding='utf-8') as f:
+                return json.loads(f.read())
+        system_message = SystemMessage(content=req_prompt)
+        video_message = HumanMessage(content=contents)
+        asr_message = HumanMessage(content=f"解说内容：{asr_text}")
+        result = self.model.invoke([system_message, video_message, asr_message]).content
+        try:
+            result_json = json.loads(result)
+        except json.JSONDecodeError:
+            result_json = json.loads(result.replace("```json", "").replace("```", ""))
+        if cache_path is not None:
+            os.makedirs(Path(cache_path).parent, exist_ok=True)
+            with open(cache_path, 'w', encoding='utf-8') as f:
+                f.write(result)
+        return result_json