A dataset for bounding box prediction in underwater environments of the Aqua-family of hexapod robots.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
AQUA_DATA.zip.001
AQUA_DATA.zip.002
AQUA_DATA.zip.003
AQUA_DATA.zip.004
LICENSE
README.md

README.md

AquaBoxDataset

A dataset for bounding box prediction in underwater environments of the Aqua-family of hexapod robots. This dataset has hand-made annotations of bounding boxes containing the hexapod. Sources range from GoPro cameras to various on-board cameras from a tailing robot.

The dataset is made available under the releases section as two .zip files: VALID.zip and TRAINING.zip.

The TRAINING.zip file contains data from 2015 and 2016 collected near the reef outside McGill Bellairs Research Institute in Barbados. VALID.zip contains data from 2017.

Directory structure of the releases:

VALID/TRAINING
    -->  (capture session)
        --> img (raw images)
            --> NNNN.jpg
        --> yolo_out (pre-computed yolo features/outputs) 
            --> NNNN.npy
        groundtruth_rect.txt (ground truth bounding box values)
        annotations.pkl (original raw annotations file)

Descriptions:

yolo_out:

pre-computed yolo features of size 1080 with the last 6 values being
[class_confidence, x_center, y_center, width, height, confidence]
of the highest confidence yolo bounding box prediction all normalized
by the image width and height respectively

groundtruth_rect.txt:

ground truth hand-made annotations of format
[x_center, y_center, width, height] of the bounding box,
all normalized by the image width and height respectively

NOTE: THIS DATASET WILL NOT BE RELEASED UNTIL January 15, 2018 DUE TO INTELLECTUAL PROPERTY RIGHTS RESOLUTION AT THAT DATE. PLEASE CONTACT THE AUTHOR IF THERE IS A NEED FOR IT BEFORE THAT TIME.

If you use this data or related work, please cite:

@INPROCEEDINGS{shkurti2017aqua,
title={Underwater Multi-Robot Convoying using Visual Tracking by Detection},
author={Shkurti, Florian and Chang, Wei-Di and Henderson, Peter and Islam, Md Jahidul and Camilo Gamboa Higuera, Juan and Li, Jimmy and Manderson, Travis and Xu, Anqi and Dudek, Gregory and Sattar, Junaed},
Booktitle = {Proc. of The IEEE International Conference on Intelligent Robots and Systems (IROS)},
Year = {2017}
}