Pamir_Visual_Inertial_Dataset

The data are located at https://huggingface.co/datasets/afrl-uw/Pamir_Visual-Inertial_Data

This dataset contains visual inertial data for three experiments at the Pamir shipwreck, Barbados.

Three dives were performed at the Pamir shipwreck with durations of 55 (2024), 51 and 51 (2025) minutes respectively, while data were recorded for 39, 44, and 45 minutes respectively. In the first dive the cameras were attached to a sensor rig. In the second and third dives two calibration targets (4-by-2 and 5-by-4) were deployed in two different locations and stayed fixed for the duration of both dives.

The directory structure is as follows:

• Videos
  -- 2024_LeftOnRig/
  -- 2024_RightOnRig/
  -- 2025_Pamir_1/
  -- 2025_Pamir_2/
• Ros2Bags
  -- 2024_LeftOnRig/
  -- 2024_RightOnRig/
  -- 2025_Pamir_1/
  -- 2025_Pamir_2/
• WaterDepth
  -- 2024_Perdix.csv/
  -- 2025_1_Perdix.csv/
  -- 2025_2_Perdix.csv/
• Configurations
  -- Target_4x2.yaml
  -- Target_5x4.yaml
  -- GoProCamIMU.yaml

Citation

If find this dataset useful, please cite:

@InProceedings{ChatzispyrouICRA2026,
  author = 	 {Michalis Chatzispyrou and Luke Horgan and Hyunkil Hwang and Harish Sathishchandra and Chinmay Burgul and Monika Roznere and Alberto {Quattrini Li} and Philippos Mordohai and Ioannis Rekleitis},
  title = 	 {{Mapping Pamir: Multi-Session Visual/Inertial SLAM and 3D Reconstruction of an Underwater Shipwreck}},
  booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
  year = 	 2026,
  month = 	 {Jun.},
  address =  {Vienna, Austria}
  }