PDF Paper ArXiv arXiv Youtube Video Github
Plane

One Flight Over the Gap

A Survey from Perspective to Panoramic Vision




Xin Lin†, Xian Ge†, Dizhe Zhang*, Zhaoliang Wan, Xianshun Wang,
Xiangtai Li, Wenjie Jiang, Bo Du, Dacheng Tao, Ming-Hsuan Yang, Lu Qi
† Equal Contribution. * Project Leader.









Insta360 Logo

PDF Paper ArXiv arXiv Code

Technical Summary Video

Panoramic vision provides holistic 360 perception, supporting applications in VR, autonomous driving, and embodied robotics. Unlike conventional perspective images, panoramic imagery exhibits unique geometric distortions, non-uniform spatial sampling, and boundary continuity, creating a substantial domain gap that hinders direct transfer of existing vision models.

Foundational Knowledge

Panorama stitching involves classical ISP preprocessing (demosaicing, denoising, correction), keypoint-based data association, geometric alignment (RANSAC + homography), and image blending. For 360° images, spherical projection serves as the base representation, from which planar forms are derived: (b) equirectangular (longitude–latitude), (c) cubemap (six 90° faces), (d) icosahedron (near-uniform sampling), (e) tangent (local planar mapping), and (f) panini (vertical preservation with horizontal compression). Projection figure

Three Gaps

From spherical image to Panoramic ERP and perspective image, ERP preserves a complete field of view compared to perspective images, but introduces three major domain gaps: (1) geometric distortion, (2) non-uniform spatial sampling, and (3) boundary continuity.. Gaps figure

Methodological Analysis

The best part? Cross-method and Cross-task analysis across four major directions: visual quality enhancement, visual understanding, multimodal fusion, and visual generation, covering 20+ tasks and nearly 300 papers. Method_1 figure Method_2 figure

Future Directions

We outline several future directions, including building larger and more diverse datasets, developing foundational, multimodal, and generation models, and extending to broader downstream applications such as embodied intelligence, autonomous driving, and immersive media. Future Direction figure

Acknowledgment

We would like to express our sincere gratitude to Yunning Peng, Haoran Feng, Shi Luo, Ruihua Lu for their valuable contributions that greatly improved the quality of this paper. We also gratefully acknowledge the generous support of Antigravity Team and Insta360 Research Team, whose assistance in various aspects made this work possible.

Citation

@article{lin2025panorama,
  title={One Flight Over the Gap: A Survey from Perspective to Panoramic Vision},
  author={Lin, Xin and Ge, Xian and Zhang, Dizhe and Wan, Zhaoliang and Wang, Xianshun and Li, Xiangtai and Jiang, Wenjie and Du, Bo and Tao, Dacheng and Yang, Ming-Hsuan and Qi, Lu},
  journal={arXiv},
  year={2025}
}