Multi-Environment Underwater Video Summarization Dataset (MEUVS)

The Multi-Environment Underwater Video Summarization Dataset (MEUVS) is a benchmark dataset designed to advance research in underwater video understanding and summarization. The dataset comprises 69 high-definition underwater video sequences collected from deep-sea, surface ocean, and freshwater environments, capturing a wide range of visual conditions and aquatic activities. MEUVS contains 39,974 frames with 246 expert-annotated temporal segments, enabling the development and evaluation of AI-based video summarization models. The dataset incorporates real-world underwater challenges, including turbidity, light attenuation, marine snow, sediment noise, illumination variations, and complex fish behaviors. By providing diverse environmental scenarios and frame-level temporal annotations, MEUVS serves as a valuable resource for underwater computer vision, video summarization, event detection, and intelligent aquatic monitoring applications. Dataset Structure: annotations/: Ground-truth boundary CSV files (train_intervals.csv, test_intervals.csv). fish_metadata.csv: Metadata for all video sequences. frames/: Extracted image frames organized into train and test sets. videos/: Raw .mp4 videos separated into train and test folders.