Generate facial attribute and emotion descriptions for each person tracked in a video using a video-to-text model (VideoLLaMA3). This operator crops face-centric video segments and generates detailed descriptions of facial expressions and emotions. It must be operated after video_human_tracks_extraction_mapper.
使用视频转文本模型（VideoLLaMA3）为视频中每个跟踪的人物生成面部属性和情感描述。此算子裁剪以面部为中心的视频片段，并生成面部表情和情感的详细描述。它必须在 video_human_tracks_extraction_mapper 之后运行。
Type 算子类型: mapper
Tags 标签: gpu, hf, video
| name 参数名 | type 类型 | default 默认值 | desc 说明 |
|---|---|---|---|
face_track_query |
<class 'str'> | 'Please describe the person's facial expression, tell me the person's emotion through the video, ...' |
prompt for describing facial attributes and emotions. |
trust_remote_code |
<class 'bool'> | False |
whether to trust the remote code of HF models. |
cropping_face_video_temp_path |
<class 'str'> | './temp_video_path' |
path for temporary face-cropped video storage. |
video_describe_model_path |
<class 'str'> | 'DAMO-NLP-SG/VideoLLaMA3-7B' |
path to the VideoLLaMA3 model. |
video_facetrack_attribute_emotion |
<class 'str'> | 'video_facetrack_attribute_emotion' |
field name to store the face attribute emotion results. |
args |
'' |
extra args | |
kwargs |
'' |
extra args |