Computer Vision and Deep Learning for Coral Ecology
While coral reefs occupy less than 0.2% of the ocean bottom, they harbor 25% of the marine life and protect shorelines occupied by 200 million people. Globally, coral reefs are in rapid decline, and scientists, reef managers, and policy makers need accurate data about the state (species, coverage) and change of individual reefs. Photographic surveys are the primary source of raw data, and it is becoming cheaper and easier to acquire massive quantities of digital images with autonomous robotic underwater vehicles. Yet even for surveys conducted by scuba divers, manual image analysis is error prone and has become a time consuming and costly bottleneck. Our work using computer vision and machine learning is breaking this bottleneck.
Research from the UCSD Computer Vision for Coral Ecology Project has lead to new cameras, algorithms, software, and services. We have developed a series of methods for automatic annotation of benthic images and created CoralNet (coralnet.ucsd.edu) as a public, open source, hosted tool for scientists to upload their photo surveys and perform annotation using deep nets trained on their data. Through a comparative study published on PLOS ONE, CoralNet is as accurate at estimating coral coverage as human experts. To date CoralNet is being used by a thousand scientists who have uploaded over 750,000 images with over 27 million expert annotations. To increase accuracy further, we looked for stronger signals for classifying corals. Using our wide field fluorescence imaging (Fluoris) camera, the error rate of deep nets applied to registered 5-channel RGB-Fluoro images decreased by 22%.
David Kriegman is a Professor of Computer Science & Engineering at the University of California, San Diego. His core research is in computer vision and machine learning, which he has applied to face recognition, robotics, coral ecology, medical imaging, microscopy, and computer graphics. Kriegman has started two companies that leveraged his research. In 2007 he founded Photometria (now operating as Sight Commerce), which uses photorealistic virtual makeover and accessory try-on to help media, brand and retail customers meet their business objectives. He co-founded KBVT, a face recognition company, which Dropbox acquired in 2014. Until his recent return to UCSD, he led the Dropbox machine learning team.
Kriegman has received best paper awards at the three major computer vision conferences: IEEE Conference on Computer Vision and Pattern Recognition (CVPR), the European Conference on Computer Vision (ECCV), and the International Conference on Computer Vision (ICCV) as well as the Paper of the Year Award from the Journal of Structural Biology. He was the Editor-in-Chief of the IEEE Trans. on Pattern Recognition and Machine Intelligence and is a Fellow of the IEEE.