The video shows rendered depth, rendered images, and a 3D Gaussian map, followed by side-by-side views labelled “Ours” and “MVS-GS”. Examples include building exteriors and indoor scenes, with camera markers illustrating novel-view selection.
MVS-GS — Visual overview
The video introduces an RGB-to-3D mapping pipeline: camera-pose estimation, multi-view stereo depth estimation, depth refinement, and Gaussian mapping. It then shows reconstructed aerial building scenes and side-by-side rendering comparisons labelled “Ours” and “Photo-SLAM”.