Marine and coral visual understanding

Abstract

Oceans, covering more than 70% surfaces of our blue planets are less explored by the whole computer vision community. Furthermore, coral reefs represent one of the most diverse and productive ecosystems on the planet, providing habitat and shelter for a vast range of marine species. However, the scarcity of labeled data and the limited diversity in existing datasets are considered the most significant hindering issues. In this thesis, we present a novel and comprehensive dataset called MarineMaid specifically designed for marine monitoring and understanding, including a wide spectrum of marine creatures. Based on the essential requirements of the marine research community, we adopt object detection and vision-language understanding as our two fundamental tasks. Additionally, we propose HKCoral, a large-scale dataset with dense semantic annotations for coral reefs, focusing on coral growth forms and capturing real-world variability. Our work addresses key challenges in marine and coral reef analysis, providing valuable tools for more efficient and accurate analysis. We also showcase practical applications, such as an AI-based marine video analysis platform and a marine debris detection system, aiming to integrate advanced technologies into real-world marine research and conservation efforts.

Date of Award	2024
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Sai Kit YEUNG (Supervisor)