ReefNet Dataset Enables AI-Driven Coral Reef Monitoring with 925K Annotations

ReefNet, a newly launched extensive public dataset, responds to the urgent requirement for automated monitoring of coral reefs, which are rapidly deteriorating due to human-induced factors such as climate change. This dataset compiles images from 76 selected CoralNet sources along with an additional site in Al-Wajh, Red Sea, amassing around 925,000 annotations at the genus level for hard corals. These annotations align with the standardized taxonomy of the World Register of Marine Species (WoRMS), promoting taxonomic uniformity across various research sites. Researchers established a high-confidence benchmark subset, achieving 92% expert agreement across 39 hard-coral label categories through rigorous verification. The lack of extensive datasets with detailed, consistent labels has hindered advancements in coral analysis, making ReefNet a pivotal step towards scalable monitoring of vulnerable marine environments.