NVS-HO
收藏资源简介:
NVS-HO是首个专注于手持物体新视角合成的基准数据集,由博洛尼亚大学、塞维利亚大学和Eyecan.ai联合创建。数据集包含67个常见物体的RGB序列,每个物体包含手持序列(HS)和固定板序列(BS),分别用于训练和评估。数据采集使用OAK-D Lite相机,分辨率1080×1920,每个物体约100-200帧,覆盖360°视角。数据集通过ChArUco板标记提供精确的相机姿态,旨在推动基于RGB的手持物体新视角合成研究,应用于机器人、增强现实和3D重建等领域。
NVS-HO is the first benchmark dataset dedicated to novel view synthesis of handheld objects, jointly created by the University of Bologna, the University of Seville, and Eyecan.ai. The dataset includes RGB sequences of 67 common everyday objects, where each object has a handheld sequence (HS) and a fixed-board sequence (BS), which are used for training and evaluation respectively. The data was collected using an OAK-D Lite camera with a resolution of 1080×1920. Each object has approximately 100 to 200 frames and covers a 360° field of view. The dataset provides precise camera poses via ChArUco board markers, aiming to advance RGB-based novel view synthesis research for handheld objects, with applications in robotics, augmented reality, 3D reconstruction, and other related fields.

- 1NVS-HO: A Benchmark for Novel View Synthesis of Handheld Objects博洛尼亚大学; 塞维利亚大学; Eyecan.ai · 2026年



