Seminar: Unsupervised Domain Adaptation of Underwater Images, 2nd June, 1PM

When: Thursday 2nd of June, 1PM AEDT

Where: This seminar will be partially presented at the Rose Street Seminar area (J04) and partially online via Zoom, RSVP here.

Speaker: Heather Doig

Title: Unsupervised Domain Adaptation of Underwater Images

Abstract:

Thousands of high quality images of the underwater environment are being captured through the ACFR’s marine robotics program.  Machine learning tasks such as classification and object identification can be used to automate image analysis but these models rely on ground truth annotations from experts for training.  Only a small percentage of images have been annotated which results in machine learning models that do not transfer well to images taken from different robotic platforms, regions or water and lighting conditions.  This seminar will present how unsupervised domain adaptation can be used to adapt machine learning models that have been trained on one source domain to a new target domain to reduce the need to annotate images from the target domain.  Two approaches will be reviewed that adapt the domain at a representation layer and at an image level.  Intermediate results  will be discussed and how future research will enhance these techniques for domain shifts between different sets of underwater images.

Bio:

Heather Doig is a second year PhD student with the Marine Systems Group at the Australian Centre for Field Robotics.  She is working on improving automated observations of the underwater environment by applying current machine learning techniques to underwater images. Heather has a Bachelor of Engineering – Mechanical and Manufacturing (Honours) from the University of Melbourne and more recently a Masters of Data Science from the University of Sydney.

Contacts

Sydney Institute for Robotics and Intelligent Systems
info@acfr.usyd.edu.au