Initialize the long video cotqa pipeline with API models. Args: backend: Video backend for info extraction (open

Initialize the VideoFilteredClipGenerator operator. Args: backend: Video backend for info extraction (opencv, to

Args: use_video_in_qa: 是否在生成 QA 时输入视频。 True: 同时使用 caption 和视频生成问题 False:

db_config: { 'host': 'localhost', 'port': 9000, 'user': 'default', 'password': '',

初始化 LazyLoader 模块。 :param name: 模块名称 :param import_structure: 定义类名到文件路径的映射字典

Initialize OCR analysis operator Args: figure_root: Directory where frames were extracted input_vide

Initialize the evaluator. Args: use_stemmer: Whether to use stemmer in ROUGE score calculation

Initialize the VisionDependentFilter operator. Args: rollout: Model for generating responses (will run without v

Initialize VideoLuminanceFilter operator. Args: figure_root: Directory where frames were extracted i

初始化 Args: min_score: 保留样本的最小运动分数 max_score: 保留样本的最大运动分数 sampling_fps: 光流计算的采样帧率（帧/秒）

Initialize the CLIPScoreEval operator. Args: model_name: CLIP model name from HuggingFace batch_size

初始化分数过滤算子 Args: min_score: 最小分数阈值（包含），低于此分数的行将被标记为 False max_score: 最大分数阈值（包含），高于此分数的行将被标记为 False

Initialize the DataTailorEval operator. Args: model_name: Qwen2-VL model name from HuggingFace clust

Initialize the KNNSimilarityFilter operator. Args: model_name: CLIP model name from HuggingFace k_ne

Initialize the FailRateFilter operator. Args: rollout: Vision-language model for generating responses

初始化 Args: min_width: 最小水平分辨率（宽度） max_width: 最大水平分辨率（宽度） min_height: 最小垂直分辨率（高度）

初始化数据集过滤算子 Args: 各种过滤条件参数，与filter_clips函数对应 seed: 随机种子，用于保证可重复性 strict_mode: 是否严格模式，True

Initialize VideoAestheticFilter operator. Args: figure_root: Directory where frames were extracted i

Initialize VideoOCRFilter operator. Args: figure_root: Directory where frames were extracted input_v

Initialize the Multi-Modal Curator Filter.

Initialize the DeduplicateFilter operator. Args: model_name: CLIP model name from HuggingFace thresh

Initialize VideoMergedCaptionGenerator operator. Args: caption_key: Column name for caption text id_