BT3.8k: A Benchmark Dataset for Monitoring Bioluminescent Algal Blooms (Blue Tears) in Coastal Zones

This dataset, BT3.8k, is the first publicly available, real-world dataset specifically designed for the detection and segmentation of bioluminescent algal blooms (locally known as "Blue Tears"). It contains 3,827 fully annotated images extracted from 241 user-uploaded social media videos from platforms like TikTok, YouTube, and WeChat. The dataset covers diverse real-world conditions, including varying lighting (night, dusk, sunset), different perspectives (ground-level, aerial/drone, boat-based), and complex coastal environments. Annotations were generated through a rigorous two-stage AI-assisted workflow (SAM2 and YOLOv11) and validated by domain experts with a high inter-annotator agreement (Dice: 0.95).