Embedded Files
Above is an example of a fully annotated image from this project. 200 points are dropped on each photograph and the program on CoralNet attempts to identify what is at each point. As of now, the AI is not very accurate and the bulk of the ID has to be manually input. A blue point signifies that whatever it has been dropped on has been verified.
Starting in Fall 2025, I have been working in the Sebens lab in the University of Washington’s Department of Biology training a machine learning model on species identification through CoralNet. My work was part of a decades-long study mapping underwater benthic cover in the Gulf of Maine and how it changes across time. I also assisted in creating an Excel dataset of mobile species (sea stars, urchins, etc.) in the photos.
When I began, all the species I was identifying were unfamiliar to me, so I had to quickly learn species ID from scratch using a packet of photographs of relevant species present in the quadrats. In that period, I learned what to look for when identifying species using a reference key and photographs of varying quality.
Most of my work was self-supervised, so I was left to my own devices and trusted to be productive. To meet that demand, I had to get much better at self-motivation and set benchmarks for myself to ensure I was an active contributor each time I was in the lab. I find that structure is a major way I keep myself productive, so learning to create my own structure was a big deal.
Page updated
Google Sites
Report abuse