Heather Doig

I am a PhD student researching the use of machine learning to interpret images from marine robotic applications at the Australian Centre for Robotics, The University of Sydney.  My current research is focused on creating training frameworks for detecting benthic species with low numbers of annotations through improved training and synthetic image generation.

EmailLinkedIn
Snorkeller in the water

Current Projects

Generating synthetic training images for object detection of benthic species with low numbers of annotations.

I've developed a handfish detector for identifying endangered handfish in images of the seafloor

Publications

Improving benthic classification using resolution scaling and SymmNet Unsupervised Domain Adaptation

Accepted to ICRA 2023, London.

Paper        Dataset        Code

Block diagram of framework

Adversarial and generative autoencoders for unsupervised clustering of benthic images

Accepted to ACRA 2020.

Paper Video

Underwater robot