CHKI-Video
收藏资源简介:
CHKI-Video数据集是由南洋理工大学S-Lab创建的一个高质量视频数据集,用于可控的人体中心关键帧插值(CHKI)算法的系统评估。数据集包含了2614个高质量视频片段,总共有超过18万个帧,这些视频来源于SportsSlomo和Pexels网站。每个帧都被精心标注了边界框、分割掩码、2D人体姿态和SMPL-X参数。数据集旨在解决人体运动插值中复杂运动和运动模糊的问题,提供丰富的空间和结构线索,以实现更加精确和逼真的人体姿态插值。
The CHKI-Video dataset is a high-quality video dataset created by S-Lab at Nanyang Technological University (NTU), designed for systematic evaluation of controllable human-centric keyframe interpolation (CHKI) algorithms. It consists of 2,614 high-quality video clips totaling over 180,000 frames, which are sourced from SportsSlomo and Pexels websites. Each frame is meticulously annotated with bounding boxes, segmentation masks, 2D human poses, and SMPL-X parameters. This dataset aims to address the challenges of complex motions and motion blur in human motion interpolation, providing rich spatial and structural cues to enable more accurate and realistic human pose interpolation.
- 1Controllable Human-centric Keyframe Interpolation with Generative Prior南洋理工大学S-Lab · 2025年



