SODD – Subaquatic Object Detection Dataset

The SODD dataset consists of 3168 images of underwater objects consistent with underwater installations, including the following categories: propeller, pipe, pipe_type2, net, red_fin, qr_codes. The total number of annotated objects are 8934, with the following distribution among categories: propeller (1092 instances), pipe (2008), pipe_type2 (886), red_fin (760), net (1556), qr_codes (2632).  

The images were acquired from a collection of videos (mp4 format, with HD resolution, and 16 FPS). The videos were acquired in an indoor pool using a Blue Robotics low-light USB camera. The vehicle used is a BlueROV2 from Blue Robotics with the heavy configuration retrofit kit, providing full actuation in 6 degrees of freedom.

More details in the documentation enclosed in the file SODD_Documentation. 

Acknowledgements

We would like to thank Professors Damiano Varagnolo and Annette Stahl from the Norwegian University for Science and Technology for the valuable advice during the planning phase of the data collection.