AquaMVS Example Dataset: Multi-View Underwater Images for 3d Reconstruction

Multi-view image dataset for testing and demonstrating AquaMVS, a Python library for underwater surface reconstruction through refractive interfaces. Contains synchronized frames from 13 above-water cameras (12 ring + 1 center) arranged in a circular rig looking down through a flat water surface at a population of Aulonocara sp. cichlids. 

The dataset includes both raw (unfiltered) and temporally median-filtered image sets at 5 time-points, along with a refractive camera calibration (produced by AquaCal) and an example AquaMVS configuration file. Temporal filtering removes transient objects (fish, bubbles, debris) to improve reconstruction of the static substrate surface.

Collected at Georgia Institute of Technology, Atlanta, Georgia on 2026-02-18. 1600x1200 resolution, 30 fps source video, 13 cameras at ~0.64 m ring radius, ~1 m water-to-rig distance.

For usage instructions, see the included README.md or the AquaMVS documentation at https://aquamvs.readthedocs.io/.