Description
Reconstructing 3D scenes from photos or video usually needs global optimization that is slow on long sequences. Spann3R introduces spatial memory for incremental, frame-by-frame dense 3D reconstruction without camera parameters.
It was a 3DV 2025 award candidate, with v1.01 checkpoints released.
Spatial memory:Accumulated scene info.
Incremental:Frame by frame.
Calibration-free:No camera parameters.
It was a 3DV 2025 award candidate, with v1.01 checkpoints released.
Features
Spatial memory:Accumulated scene info.
Incremental:Frame by frame.
Calibration-free:No camera parameters.
