Merge branch 'master' of https://github.com/MaiBot-Plus/MaiMbot-Pro-Max
This commit is contained in:
雅诺狐
@@ -130,6 +130,8 @@ class VideoAnalyzer:
|
||||
# 新增的线程池配置
|
||||
self.use_multiprocessing = getattr(config, 'use_multiprocessing', True)
|
||||
self.max_workers = getattr(config, 'max_workers', 2)
|
||||
self.frame_extraction_mode = getattr(config, 'frame_extraction_mode', 'fixed_number')
|
||||
self.frame_interval_seconds = getattr(config, 'frame_interval_seconds', 2.0)
|
||||
|
||||
# 将配置文件中的模式映射到内部使用的模式名称
|
||||
config_mode = config.analysis_mode
|
||||
@@ -163,6 +165,8 @@ class VideoAnalyzer:
|
||||
self.enable_frame_timing = True
|
||||
self.use_multiprocessing = True # 默认启用线程池
|
||||
self.max_workers = 2 # 默认最大2个线程
|
||||
self.frame_extraction_mode = "fixed_number"
|
||||
self.frame_interval_seconds = 2.0
|
||||
self.batch_analysis_prompt = """请分析这个视频的内容。这些图片是从视频中按时间顺序提取的关键帧。
|
||||
|
||||
请提供详细的分析，包括：
|
||||
@@ -314,6 +318,8 @@ class VideoAnalyzer:
|
||||
async def _extract_frames_fallback(self, video_path: str) -> List[Tuple[str, float]]:
|
||||
"""帧提取的降级方法 - 原始异步版本"""
|
||||
frames = []
|
||||
frame_count = 0
|
||||
extracted_count = 0
|
||||
cap = cv2.VideoCapture(video_path)
|
||||
fps = cap.get(cv2.CAP_PROP_FPS)
|
||||
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
||||
@@ -321,61 +327,97 @@ class VideoAnalyzer:
|
||||
|
||||
logger.info(f"视频信息: {total_frames}帧, {fps:.2f}FPS, {duration:.2f}秒")
|
||||
|
||||
# 使用numpy优化帧间隔计算
|
||||
if duration > 0:
|
||||
frame_interval = max(1, int(duration / self.max_frames * fps))
|
||||
else:
|
||||
frame_interval = 30 # 默认间隔
|
||||
|
||||
logger.info(f"计算得出帧间隔: {frame_interval} (将提取约{min(self.max_frames, total_frames // frame_interval + 1)}帧)")
|
||||
|
||||
# 使用numpy计算目标帧位置
|
||||
target_frames = np.arange(0, min(self.max_frames, total_frames // frame_interval + 1)) * frame_interval
|
||||
target_frames = target_frames[target_frames < total_frames].astype(int)
|
||||
|
||||
extracted_count = 0
|
||||
|
||||
for target_frame in target_frames:
|
||||
# 跳转到目标帧
|
||||
cap.set(cv2.CAP_PROP_POS_FRAMES, target_frame)
|
||||
ret, frame = cap.read()
|
||||
if not ret:
|
||||
continue
|
||||
|
||||
# 使用numpy优化图像处理
|
||||
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
||||
if self.frame_extraction_mode == "time_interval":
|
||||
# 新模式：按时间间隔抽帧
|
||||
time_interval = self.frame_interval_seconds
|
||||
next_frame_time = 0.0
|
||||
|
||||
# 转换为PIL图像并使用numpy进行尺寸计算
|
||||
height, width = frame_rgb.shape[:2]
|
||||
max_dim = max(height, width)
|
||||
|
||||
if max_dim > self.max_image_size:
|
||||
# 使用numpy计算缩放比例
|
||||
ratio = self.max_image_size / max_dim
|
||||
new_width = int(width * ratio)
|
||||
new_height = int(height * ratio)
|
||||
while cap.isOpened():
|
||||
ret, frame = cap.read()
|
||||
if not ret:
|
||||
break
|
||||
|
||||
# 使用opencv进行高效缩放
|
||||
frame_resized = cv2.resize(frame_rgb, (new_width, new_height), interpolation=cv2.INTER_LANCZOS4)
|
||||
pil_image = Image.fromarray(frame_resized)
|
||||
current_time = cap.get(cv2.CAP_PROP_POS_MSEC) / 1000.0
|
||||
|
||||
if current_time >= next_frame_time:
|
||||
# 转换为PIL图像并压缩
|
||||
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
||||
pil_image = Image.fromarray(frame_rgb)
|
||||
|
||||
# 调整图像大小
|
||||
if max(pil_image.size) > self.max_image_size:
|
||||
ratio = self.max_image_size / max(pil_image.size)
|
||||
new_size = tuple(int(dim * ratio) for dim in pil_image.size)
|
||||
pil_image = pil_image.resize(new_size, Image.Resampling.LANCZOS)
|
||||
|
||||
# 转换为base64
|
||||
buffer = io.BytesIO()
|
||||
pil_image.save(buffer, format='JPEG', quality=self.frame_quality)
|
||||
frame_base64 = base64.b64encode(buffer.getvalue()).decode('utf-8')
|
||||
|
||||
frames.append((frame_base64, current_time))
|
||||
extracted_count += 1
|
||||
|
||||
logger.debug(f"提取第{extracted_count}帧 (时间: {current_time:.2f}s)")
|
||||
|
||||
next_frame_time += time_interval
|
||||
else:
|
||||
# 使用numpy优化帧间隔计算
|
||||
if duration > 0:
|
||||
frame_interval = max(1, int(duration / self.max_frames * fps))
|
||||
else:
|
||||
pil_image = Image.fromarray(frame_rgb)
|
||||
frame_interval = 30 # 默认间隔
|
||||
|
||||
logger.info(f"计算得出帧间隔: {frame_interval} (将提取约{min(self.max_frames, total_frames // frame_interval + 1)}帧)")
|
||||
|
||||
# 使用numpy计算目标帧位置
|
||||
target_frames = np.arange(0, min(self.max_frames, total_frames // frame_interval + 1)) * frame_interval
|
||||
target_frames = target_frames[target_frames < total_frames].astype(int)
|
||||
|
||||
# 转换为base64
|
||||
buffer = io.BytesIO()
|
||||
pil_image.save(buffer, format='JPEG', quality=self.frame_quality)
|
||||
frame_base64 = base64.b64encode(buffer.getvalue()).decode('utf-8')
|
||||
|
||||
# 计算时间戳
|
||||
timestamp = target_frame / fps if fps > 0 else 0
|
||||
frames.append((frame_base64, timestamp))
|
||||
extracted_count += 1
|
||||
|
||||
logger.debug(f"提取第{extracted_count}帧 (时间: {timestamp:.2f}s, 帧号: {target_frame})")
|
||||
|
||||
# 每提取一帧让步一次
|
||||
await asyncio.sleep(0.001)
|
||||
extracted_count = 0
|
||||
|
||||
for target_frame in target_frames:
|
||||
# 跳转到目标帧
|
||||
cap.set(cv2.CAP_PROP_POS_FRAMES, target_frame)
|
||||
ret, frame = cap.read()
|
||||
if not ret:
|
||||
continue
|
||||
|
||||
# 使用numpy优化图像处理
|
||||
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
||||
|
||||
# 转换为PIL图像并使用numpy进行尺寸计算
|
||||
height, width = frame_rgb.shape[:2]
|
||||
max_dim = max(height, width)
|
||||
|
||||
if max_dim > self.max_image_size:
|
||||
# 使用numpy计算缩放比例
|
||||
ratio = self.max_image_size / max_dim
|
||||
new_width = int(width * ratio)
|
||||
new_height = int(height * ratio)
|
||||
|
||||
# 使用opencv进行高效缩放
|
||||
frame_resized = cv2.resize(frame_rgb, (new_width, new_height), interpolation=cv2.INTER_LANCZOS4)
|
||||
pil_image = Image.fromarray(frame_resized)
|
||||
else:
|
||||
pil_image = Image.fromarray(frame_rgb)
|
||||
|
||||
# 转换为base64
|
||||
buffer = io.BytesIO()
|
||||
pil_image.save(buffer, format='JPEG', quality=self.frame_quality)
|
||||
frame_base64 = base64.b64encode(buffer.getvalue()).decode('utf-8')
|
||||
|
||||
# 计算时间戳
|
||||
timestamp = target_frame / fps if fps > 0 else 0
|
||||
frames.append((frame_base64, timestamp))
|
||||
extracted_count += 1
|
||||
|
||||
logger.debug(f"提取第{extracted_count}帧 (时间: {timestamp:.2f}s, 帧号: {target_frame})")
|
||||
|
||||
# 每提取一帧让步一次
|
||||
await asyncio.sleep(0.001)
|
||||
|
||||
cap.release()
|
||||
logger.info(f"✅ 成功提取{len(frames)}帧")
|
||||
return frames
|
||||
|
||||
Reference in New Issue
Block a user