AquaMVS Example Dataset 2: Multi-View Underwater Images for Ground Truth Analysis

 This archive contains the complete input data and derived analysis artifacts used to validate the geometric accuracy
  of the AquaMVS underwater multi-view stereo (MVS) reconstruction pipeline against independent ChArUco
  calibration-board ground truth. The validation is non-circular: ground-truth corner geometry is recovered from the
  calibration boards independently of the MVS reconstruction it is used to assess.

  The dataset (~8.6 GB uncompressed) is organized into two parts:

  - aquamvs_ground_truth_analysis/ — the reconstruction inputs and outputs. Includes synchronized underwater imagery
  from a 13-camera rig (10 frames per camera), per-camera calibration (calibration.json: intrinsics, extrinsics, and
  per-camera water-surface depth), the pipeline configuration (config.yaml), and the full AquaMVS reconstruction
  output/ for each frame (depth maps, multi-view consistency maps, fused point clouds, meshes, and undistorted
  images).
  - analysis_output/ — the derived ground-truth validation artifacts. Includes transferred ChArUco corner detections
  (corner_transfer/), an error decomposition of reconstructed vs. ground-truth corner geometry
  (error_decomposition.json), scale alignment (scale_alignment.json), scale-independent metrics (cross-camera
  agreement and board flatness consistency), and publication figures (PDF/PNG/SVG).

  The companion analysis code that consumes this data and reproduces all metrics and figures is available at:
  https://github.com/McGrathLab/AquaMVS_gtanalysis

  To reproduce the analysis, extract this archive so that the data/ directory sits at the root of the analysis
  repository.