作者大大您好，在原文中的视频特征提取部分您提到”This involves scene detection, object detection (Ren et al., 2015), face detection (Zhang et al., 2017), face tracking, and audio-visual active speaker detection (Tao et al., 2021), as described in (Zhang et al., 2022a). This process can generate more than 1,000K high-quality keyframes with speaker bounding boxes in approximately 5 days. Next, we use these annotated RoIs and employ the instance segmentation method, Mask R-CNN (He et al., 2017), pre-trained on the COCO (Lin et al., 2014) dataset to extract visual features.“，请问下之后的1000K帧说话人特征提取大概花费了几天时间呢，以及整个过程（即场景检测，目标检测，人脸跟踪，特征提取等）用到的GPU型号和数量是怎样的呢？
作者大大您好，在原文中的视频特征提取部分您提到”This involves scene detection, object detection (Ren et al., 2015), face detection (Zhang et al., 2017), face tracking, and audio-visual active speaker detection (Tao et al., 2021), as described in (Zhang et al., 2022a). This process can generate more than 1,000K high-quality keyframes with speaker bounding boxes in approximately 5 days. Next, we use these annotated RoIs and employ the instance segmentation method, Mask R-CNN (He et al., 2017), pre-trained on the COCO (Lin et al., 2014) dataset to extract visual features.“，请问下之后的1000K帧说话人特征提取大概花费了几天时间呢，以及整个过程（即场景检测，目标检测，人脸跟踪，特征提取等）用到的GPU型号和数量是怎样的呢？