Unified Benchmark for Rip Current Detection and Segmentation
RipBench is the first comprehensive, multi-task benchmark for rip current analysis, supporting classification, object detection, oriented bounding box detection, semantic segmentation, instance segmentation, and panoptic segmentation (coming soon).
The benchmark includes 300 videos (equally balanced between rip and non-rip events), totaling 291,946 frames and approximately 2 hours and 36 minutes of footage collected from diverse global coastlines. All tasks are supported with carefully curated annotations and evaluated using standard and safety-critical metrics, with a focus on the F2 score to prioritize recall in this high-risk domain.
RipBench provides a unified, balanced dataset along with baseline models and evaluation code, enabling real-world progress in automated beach monitoring and early warning systems.
The full dataset, baseline models, and results are released for review and will be made completely public soon. Stay tuned! Check the History page for timeline updates.
Contact: hidden.for.review