Every row is one scene, shown on a training view and on a held-out one, with the ground truth last and each render badged with its own score. Contributions are cumulative: Base is SAM3D's own appearance prediction, +Attn adds the visibility attention bias, +RG the rendering guidance, and +TTR the test-time refinement. The cross-method comparisons are on the qualitative results page, and the interactive orbits on the main page.
Hover a tile to magnify the same point across that scene's row, and click it to open the render full size.