UWHandles Dataset

GitHub

Dataset ZIP

UWHandles is a dataset for 6D object pose estimation in underwater fisheye images. It provides 6D pose and 2D bounding box annotations for 3 different graspable handle objects used for ROV manipulation. The dataset consists of 28 image sequences collected in natural seafloor environments with a total of 20,427 annotated frames.

Citation

If you use this dataset, we request you to cite the following work

@article{billings2020silhonet, title={SilhoNet-Fisheye: Adaptation of A ROI Based Object Pose Estimation Network to Monocular Fisheye Images}, author={Billings, Gideon and Johnson-Roberson, Matthew}, journal={arXiv preprint arXiv:2002.12415}, year={2020}}

Overview

Sample annotated sequence, showing center rectified images for visualization of the model handle projections

The dataset was annotated using the VisPose annotation tool, which can also be used to review the annotations. The sequence consistent camera poses for input to the VisPose annotation tool were generated using the ROS based TagSLAM package. Below is a sample sequence showing the AprilTag detector and TagSLAM estimated camera poses