RHAS133
收藏资源简介:
RHAS133数据集是首个用于多人物场景中基于文本引用的人体动作分割的数据集。它由133部电影组成,包含33小时的视频数据和137种细粒度的动作标注,以及为每个感兴趣的人提供文本引用。该数据集旨在解决当前人体动作分割方法无法有效处理多人物场景和未定义动作序列的问题,并作为评估相关方法的基准。RHAS133数据集的独特之处在于它结合了多人物交互、文本引用和细粒度动作标注,使其成为一个更加全面和具有挑战性的基准,以推动在多人物场景中的人体动作分割研究。
The RHAS133 dataset is the first dataset for text-referring human action segmentation in multi-person scenes. It consists of 133 movies, containing 33 hours of video data, 137 fine-grained action annotations, and text references for each person of interest. This dataset aims to address the limitation that current human action segmentation methods cannot effectively handle multi-person scenes and undefined action sequences, and serves as a benchmark for evaluating relevant approaches. What distinguishes the RHAS133 dataset is its integration of multi-person interaction, text references and fine-grained action annotations, making it a more comprehensive and challenging benchmark to advance research on human action segmentation in multi-person scenarios.

- 1HopaDIFF: Holistic-Partial Aware Fourier Conditioned Diffusion for Referring Human Action Segmentation in Multi-Person Scenarios卡尔斯鲁厄理工学院, 北京理工大学, 中国科学院自动化研究所, 湖南大学, 上海人工智能实验室, 河北科技大学 · 2025年



