Coral Segmentation - A 'Highlight' in IEEE/CVF Computer Vision and Pattern Recognition (CVPR)

Congratulations to Prof. Sai-Kit Yeung and Ms Haixin Liang (Mphil student of ISD) for getting their groundbreaking research paper selected as a 'Highlight' in the top-tier conferences -  IEEE/CVF Computer Vision and Pattern Recognition (CVPR) 2024. The research team presents the first foundation model for coral image segmentation and has hopes to turn it into a new standard tool for coral visual research.

Coral is one of the key indicators of marine research due to coral’s rich biodiversity and sensitivity to small environmental changes. However, segmenting coral is no easy task due to its irregular boundaries and degraded water quality. Therefore, underwater coral visual understanding has gained increasing attention within the computer vision community. Prof. Yeung and his team introduced CoralSCOP, the first foundation model that can segment dense coral reef automatically and has strong generalization ability to project the full image of coral reef. The model also offers user-defined tuning and sparse-to-dense conversion for precise coral statistics. To maximize its usage, the team also incorporated mask-referring segmentation and instruction-following segmentation so that both amateurs and advanced researchers could master the model.

Computer Vision and Pattern Recognition Conference (CVPR) ranks top on Research.com’s 2023 Best Computer Science Conference List. Among the 11.532 valid papers, only 2.8% of them were selected as highlights. This appreciate recognizes the innovation of ISD and Prof. Yeung’s team. The research results contribute to human capacity in underwater computer vision research and signify HKUST’s advanced ability in Marine AI research.

For more information about CoralSCOP, please visit the website: https://coralscop.hkustvgd.com.