We have often talked about how difficult it is the get data from the deep sea… but would you believe that the bottleneck to our understanding of the deep ocean, at least as far as visual data, is processing those images?

Turning a picture of the deep sea into a list of species, habitat type, sediment type etc. is a time-consuming process that requires a wide range of skilled people.

Due to time/funding constrains a lot of valuable information is lost. A team looking at a specific question will have lots of information in their data that other teams could use.

A picture is worth a thousand data points

We chat with Kakani Katija, the co-founder of FathomNet, an open-source repository for labelled deep-sea imaging data. The platform is still in beta but it is hoped that it will allow scientists to easily and usefully share their amassed data in a single and easily searchable place.

But what about that processing bottleneck? The tech-savvy listener may have noticed that a massive collection of labelled image data is exactly the sort of thing you need to train a Machine Learning or Deep Learning algorithm. Can we automate a lot of the time-consuming image processing and let the experts focus on the new and unusual stuff? It’s at this cutting edge that things get exciting and we may be at the cusp of a marine science renaissance.

We have all our usual nonsense. Alan continues to find things in Australia that want to kill him. We answer some listener questions on deep-sea fish memory and Thom and Alan’s origin story. We also hear from Don about his time on the cruise ships. A decidedly un-deep-sea segment but it could be considered ‘adventure’ on the high seas.