Gait3D
收藏资源简介:
Gait3D数据集是由杭州电子科技大学和京东探索研究院合作创建的大型3D步态识别数据集。该数据集包含4000个主题,超过25,000个序列,这些序列是从39个摄像头的非约束室内场景中提取的。Gait3D不仅提供了传统的2D轮廓和关键点,还提供了从视频帧中精确恢复的3D人体网格,包括3D姿势、形状和视角参数。这些多模态数据为步态识别研究提供了丰富的信息资源。Gait3D数据集的应用领域主要集中在解决现实世界中的步态识别问题,特别是在视角变化、遮挡和复杂环境干扰下的步态识别挑战。
The Gait3D dataset is a large-scale 3D gait recognition dataset jointly developed by Hangzhou Dianzi University and JD Explore Academy. This dataset includes 4000 subjects and over 25,000 sequences captured from unconstrained indoor scenes via 39 cameras. Beyond conventional 2D silhouettes and keypoints, Gait3D also provides precisely recovered 3D human meshes derived from video frames, encompassing 3D pose, shape and viewpoint parameters. These multimodal data offer abundant informational resources for gait recognition research. The primary application scope of the Gait3D dataset lies in addressing real-world gait recognition challenges, particularly those arising from viewpoint variations, occlusions and complex environmental disturbances.




