Swimm3R: Splatting with Medium-aware SfM for Underwater 3D Reconstruction

1 min read

Swimm3R attacks the step that fails first in murky water, structure from motion. It distils in-air geometric priors into a feed-forward backbone with a physics head that regresses underwater image formation parameters, camera poses and a restored point cloud. On top it adds Underwater Beta Splatting: Gaussian splatting with Beta primitives and scattering-aware geometric gradients for more stable geometry in water. The authors also release a Barbados underwater video dataset.

Findings

  • On the Barbados set, Underwater Beta Splatting from Swimm3R’s point clouds improves average PSNR by 1.47 dB over WaterSplatting (abstract).
  • Downstream localisation improves by 2.0 and 2.4 percentage points in RRA@15 and RTA@15.

Limits

  • Abstract only so far. One new dataset from one region.

Relevance

Moving the medium model into the pose and structure step, as well as the renderer, responds to the finding in alvareztunon-2026-gs-underwater that SfM registers nothing at high turbidity. See underwater perception and Wat3R.