Dataset: Application of SymmNet Unsupervised Domain Adaptation and Resolution Scaling for Improved Benthic Classification

Published: 2 February 2023| Version 3 | DOI: 10.17632/d2yn52n9c9.3
Contributors:
Heather Doig, Oscar Pizarro, Stefan Williams

Description

This dataset contains patch images of benthic species and physical features from two sets of AUV surveys used for machine learning research into Unsupervised Domain Adaptation. The dataset includes a README.md explaining the structure of the directories and samples of the image patches by class. The data was used in the ICRA 2023 paper "Application of SymmNet Unsupervised Domain Adaptation and Resolution Scaling for Improved Benthic Classification". The first dataset (EMR) is from Elizabeth and Middleton Reef, Australia captured in January 2020 by the AUV Sirius and the AUV Nimbus including 2658 image patches with 6 classes (5 benthic species and one physical feature). The second dataset (SHR) is from the South Hydrate Ridge in the North Eastern Pacific captured by the AUV AE2000F and the AUV Tuna-sand during the Adaptive Robotics cruise in 2018 including 949 image patches with 8 classes (5 benthic species and 3 physical features). This dataset is intended for machine learning research. The original images and manual point annotations are available on squidle.org (EMR) and soi.squidle.org (SHR). See 'Steps to reproduce' for more details. ACKNOWLEDGMENT The data from the EMR dataset was part of Australia’s Integrated Marine Observing System (IMOS) – IMOS is enabled by the National Collaborative Research Infrastructure Strategy (NCRIS). The EMR annotations were based on work undertaken for the Marine Biodiversity Hub, a collaborative partnership supported through funding from the Australian Government’s National Environmental Science Program (NESP). The images for the SHR dataset were collected during the Schmidt Ocean Institute's FK180731 Adaptive Robotics campaign, with support from the Japanese Government's Zipangu in the Ocean Strategic Innovation Program.

Files

Steps to reproduce

The image patches can be reproduced (ie cropped and resized) from the original AUV images and manual point annotations available on squidle.org (EMR dataset) and soi.squidle.org (SHR dataset). EMR Dataset on squidle.org 1. Register for a login to squidle.org 2. From 'Datasets' select Group='ACFR Marine', Datasets='EMR Elizabeth NG06 < 30m', 'EMR Middleton 30+ m', 'EMR Middleton' and Annotation Set = 'HJD_classifier_training'. 3. Download each annotation set which will include the link to the original image and the location and class of the manual point annotation. SHR dataset on soi.squidle.org 1. Register for a login to soi.squidle.org 2. From 'My Datasets' select Group/Media Collection/AnnotationSet = Public/SHR_TUNASAND_3000Sample/HJD_UDA_classification and Public/SHR_AE2000_3000Sample/HJD_UDA_classification 3. Download each annotation set which will include the link to the original image and the location and class of the point annotation.

Institutions

University of Sydney Australian Centre for Field Robotics, University of Sydney

Categories

Computer Vision, Robotics, Benthic Ecology, Transfer Learning

Licence