PP-HumanSeg
收藏资源简介:
PP-HumanSeg是由百度公司创建的大规模视频人像分割数据集,包含291个视频,来自23个不同的会议场景,总计14,117个精细标注的帧。该数据集通过在真实世界的视频会议场景中收集视频,并由专业标注人员进行像素级和视频级的标注,确保了数据的高质量和多样性。数据集的创建旨在解决视频会议中人像分割的挑战,特别是在后疫情时代,当视频会议环境更加多样化和复杂化时。PP-HumanSeg的应用领域包括但不限于视频会议背景替换、虚拟现实和增强现实等,旨在提供高质量的人像分割解决方案,以增强视频会议的交互体验。
PP-HumanSeg is a large-scale video human segmentation dataset created by Baidu. It consists of 291 videos sourced from 23 distinct conference scenarios, with a total of 14,117 finely annotated frames. This dataset is collected from real-world video conferencing scenarios, and annotated at both pixel-level and video-level by professional annotators, which ensures high data quality and diversity. The dataset is developed to address the challenges of human segmentation in video conferencing, especially in the post-pandemic era when video conferencing environments have become more diverse and complex. Its application fields include but are not limited to video conferencing background replacement, virtual reality (VR) and augmented reality (AR), aiming to provide high-quality human segmentation solutions to enhance the interactive experience of video conferencing.

- 1PP-HumanSeg: Connectivity-Aware Portrait Segmentation with a Large-Scale Teleconferencing Video Dataset百度公司 · 2021年



