HUMBI
收藏资源简介:
HUMBI数据集是由明尼苏达大学创建的大型多视角人体表达数据集,旨在促进视角特定外观和几何形状的建模。该数据集包含772个独特的主题,通过107个同步的高清摄像头捕捉,覆盖性别、种族、年龄和身体状况的多样性。数据集内容包括面部、注视、手部、身体和衣物的表达,通过3D网格模型重建高保真的人体表达,允许使用其规范图集表示视角特定的外观。HUMBI数据集的应用领域包括学习完整的人体模型重建,以及与现有视角和主题有限的人体表达数据集互补,解决复杂物理交互中的挑战,如半透明皮肤、微小皱纹和反射面料。
The HUMBI dataset is a large-scale multi-view human expression dataset developed by the University of Minnesota, aiming to advance the modeling of view-specific appearance and geometry. This dataset includes 772 unique subjects, captured via 107 synchronized high-definition cameras, and covers diversity across gender, race, age and physical conditions. The dataset encompasses expressions of the face, gaze, hands, full body and clothing. It reconstructs high-fidelity human expressions through 3D mesh models, enabling the use of its canonical atlas to represent view-specific appearances. Application scenarios of the HUMBI dataset include learning complete human model reconstruction, and complementing existing human expression datasets with limited views and subjects to address challenges in complex physical interactions, such as translucent skin, fine wrinkles and reflective fabrics.




