Datasets:

davidesapienza
/

UnderWaterNet

license: openrail
task_categories:
  - object-detection
  - robotics
tags:
  - underwater
  - 6d-pose-estimation
  - computer-vision
  - synthetic
  - unity
size_categories:
  - 10K<n<100K
pretty_name: UnderWaterNet Dataset

UnderWaterNet Dataset

Official repository for the UnderWaterNet dataset, a comprehensive collection of underwater and surface images for computer vision tasks including 2D object detection and 6D pose estimation.

Paper

Model-Based Underwater 6D Pose Estimation From RGB
Published in IEEE Xplore

Official Website

https://sapienzadavide.github.io/uwpose.github.io/index.html

Dataset Structure

Real Underwater & Surface Images (UWds)

Compressed archives of real-world images across different environmental conditions:

UW6d/asphalt.tar.gz (4.6GB) - Asphalt surface images
UW6d/drydirt.tar.gz (3.4GB) - Dry dirt surface images
UW6d/dryshadow.tar.gz (2.5GB) - Shadowed dry surface images
UW6d/drywhite.tar.gz (3.8GB) - White dry surface images
UW6d/grass.tar.gz (4.5GB) - Grass surface images
UW6d/uwLED.tar.gz (530MB) - Underwater LED-lit images
UW6d/uwshadow.tar.gz (1.9GB) - Underwater shadowed images
UW6d/w_p_occluded.tar.gz (956MB) - Water with partial occlusion

3D Object Models (UW6d)

UW6d/objects_ply/ - PLY mesh files for 6D pose estimation tasks

2D Object Detection Dataset (UW2d)

Real-world images for 2D object detection:

UW2d/cube.tar.gz - Cube object images
UW2d/cup.tar.gz - Cup object images
UW2d/hotstab.tar.gz - Hotstab object images
UW2d/jar.tar.gz - Jar object images
UW2d/multiple_objects.tar.gz - Multi-object scenes

Synthetic Datasets

Unity-generated synthetic data for training:

synthetic_2d.tar.gz - Synthetic dataset for 2D object detection (no background)
synthetic_6d.tar.gz - Synthetic dataset for 6D pose estimation (sandy background)

Tasks Supported

6D pose estimation
2D object detection

Citation

If you use this dataset in your research, please cite:

@ARTICLE{10265120,
  author={Sapienza, Davide and Govi, Elena and Aldhaheri, Sara and Bertogna, Marko and Roura, Eloy and Pairet, Èric and Verucchi, Micaela and Ardón, Paola},
  journal={IEEE Robotics and Automation Letters}, 
  title={Model-Based Underwater 6D Pose Estimation From RGB}, 
  year={2023},
  volume={8},
  number={11},
  pages={7535-7542},
  keywords={Pose estimation;Three-dimensional displays;Sensors;Solid modeling;Task analysis;Cameras;Robot sensing systems;Computer vision for manipulation;dataset;manipulation;pose estimation;underwater},
  doi={10.1109/LRA.2023.3320028}
  }

License

openrail