RipVIS

RipVIS is a large-scale video instance segmentation benchmark for detecting rip currents from real-world beach footage. It is focused on instance segmentation for precise identification of rip currents.

Minimizing false negatives and false positives

Temporal Confidence Aggregation (TCA)

TCA is a custom temporal post-processing method designed to reduce false positives and improve temporal consistency of rip current predictions. Below is a visual breakdown of its multi-stage filtering process:

Diverse Conditions and Annotation Types

RipVIS includes scenes from drones, smartphones, and fixed cams in various lighting, wave, and sand conditions. Here's a visual overview:

Global Data Sources

RipVIS includes beach footage from a variety of locations across the USA, Mexico, Costa Rica, Portugal, Italy, Greece, Romania, Sri Lanka, Australia, and New Zealand. This map shows the origin of videos in the dataset:

Paper & Citation

Download the paper (PDF)

@inproceedings{dumitriu2025ripvis,
    author    = {Dumitriu, Andrei and Tatui, Florin and Miron, Florin and Ralhan, Aakash and Ionescu, Radu Tudor and Timofte, Radu},
    title     = {RipVIS: Rip Currents Video Instance Segmentation Benchmark for Beach Monitoring and Safety},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {3427--3437}
}
        

@inproceedings{aim2025ripseg,
    title={{AIM} 2025 Rip Current Segmentation ({RipSeg})} Challenge Report,
    author={Andrei Dumitriu and Florin Miron and Florin Tatui and Radu Tudor Ionescu and Radu Timofte and Aakash Ralhan and Florin-Alexandru Vasluianu and others},
    booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
    year={2025}
}
        

@inproceedings{dumitriu2023rip,
    title="{Rip Current Segmentation: A novel benchmark and YOLOv8 baseline results}",
    author={Dumitriu, Andrei and Tatui, Florin and Miron, Florin and Ionescu, Radu Tudor and Timofte, Radu},
    booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    pages={1261--1271},
    year={2023}
}
        

Video Presentation

Watch a short presentation introducing RipVIS, its motivation, design, and key results:

Dataset Overview

184 videos · 212,328 frames
Multi-source: drones, phones, fixed cams
1 class: rip current (multi-instance)

Note: The full dataset is available on huggingface: https://huggingface.co/datasets/Irikos/RipVIS/

License & Contact

RipVIS is released for research and safety purposes only (non-commercial use).
Contact: andrei.dumitriu@uni-wuerzburg.de