Pano360 Dataset
收藏资源简介:
Pano360数据集是由华南理工大学等机构联合构建的大规模真实场景全景数据集,包含200个多样化场景的14,400张高分辨率图像,覆盖完整360°视场角。数据集涵盖弱纹理、多变光照和极端天气等挑战性条件,每张图像均标注真实相机参数。该数据集通过专业设备采集和多视角几何验证构建,旨在解决传统全景拼接方法在复杂场景下的几何失真问题,为自动驾驶、虚拟现实等领域的3D全景生成提供基准支持。
The Pano360 dataset is a large-scale real-world panoramic dataset jointly constructed by South China University of Technology and other institutions. It comprises 14,400 high-resolution images across 200 diverse scenes, covering a full 360° field of view. The dataset encompasses challenging scenarios including low-texture regions, varying illumination, and extreme weather, with each image annotated with real camera parameters. Constructed via professional equipment collection and multi-view geometry validation, this dataset aims to address the geometric distortion issue of traditional panoramic stitching methods in complex scenes, providing benchmark support for 3D panoramic generation in fields such as autonomous driving and virtual reality.
Pano360 数据集概述
数据集基本信息
- 数据集名称:Pano360: Perspective to Panoramic Vision with Geometric Consistency
- 相关论文:Zhu, Zhengdong, et al. "Pano360: Perspective to Panoramic Vision with Geometric Consistency." arXiv preprint arXiv:2603.12013 (2026).
- 许可证:MIT
- 访问地址:https://huggingface.co/datasets/DongZhi/Pano360
数据集内容与结构
数据集包含四个不同的场景类别:
- Scene(a):旅游场景
- Scene(b):体育场景
- Scene(c):特殊光照场景
- Scene(d):无监督的野外场景
数据组织方式
每个场景目录下包含多个子场景(例如,Scene(a)包含从0到165索引的166个子场景)。每个子场景目录下包含多个焦距设置(例如001、002、003)。在每个焦距目录中,具体包含:
cameras.json:地面真实相机参数文件images/:包含恰好24个图像帧的目录
数据用途
该数据集旨在支持从透视图像到全景视觉的几何一致性研究。




