Panoramic 3D Reconstruction

Panoramic 3D reconstruction estimates scene geometry and camera motion from 360-degree images or video, enabling immersive viewing, mapping, and simulation. Methods adapt multi-view stereo, depth and pose prediction, SLAM, and Gaussian splatting to spherical camera geometry, while feed-forward models and diffusion-based view completion help when captures are sparse. Work spans single-panorama scene generation, multi-view indoor and outdoor mapping, dynamic scenes, and real-time robotic perception, with evaluation on depth, pose, point clouds, and novel views. A central challenge is balancing plausible completion of unseen areas with accurate geometry, scale, and consistency across viewpoints, especially under distortion, camera motion, and limited training data.