The BENTHOZ-2015 data set is a collection of around 10,000 images of the sea floor in coastal Australia, taken by an Autonomous Underwater Vehicle (AUV). It includes sensor data from the AUV, as well as expert annotations on over 400,000 image points. The image points are labbelled according to the CATAMI hierarchy, conforming to a taxonomy of nearly 150 classes. For more details on the data set and broader context, please read the data set descriptor on the Nature Publishing Group's Scientific Data journal (http://www.nature.com/articles/sdata201557). If you make use of the data set or this toolbox in an academic context, that is the paper to cite.
This repository consists of a growing set of tools to access, manipulate, and perform automated classification on data in the BENTHOZ-2015 data set. The primary aim is to reduce the barrier to entry in conducting research, enabling a wider range of machine learning and computer vision researchers to make progress on the automated classification of benthic imagery.
The official copy of the point metadata and labels is available on Figshare http://dx.doi.org/10.6084/m9.figshare.1524139. The simplest way to acquire the images is to follow the instructions on Squidle at https://squidle.acfr.usyd.edu.au/benthoz2015. It involves downloading a list of images, and the image downloader tool for your platform.
The data_splits folder contains labels for the training, calibration and test sets used in my PhD thesis (http://hdl.handle.net/2123/14511).