Rip Current Segmentation: A Novel Benchmark and YOLOv8 Baseline Results
Andrei Dumitriu, Florin Tatui, Florin Miron, Radu Tudor Ionescu, Radu Timofte·April 03, 2025
Summary
A novel task, rip current instance segmentation, is introduced using a comprehensive dataset with 2,466 images and 17 drone videos. The YOLOv8-nano model achieved the best results, with an mAP50 of 88.94% on validation and 81.21% macro average on the test dataset. A new benchmark was created with polygonal annotations for 2,466 images and 17 videos, offering a foundation for future research in rip current segmentation. Machine learning methods provide cost-effectiveness, scalability, and real-time detection, surpassing traditional observation methods.
Introduction
Background
Overview of rip currents and their significance
Current methods for rip current detection and limitations
Objective
Aim of introducing the rip current instance segmentation task
Importance of a comprehensive dataset for this task
Method
Data Collection
Description of the dataset (2,466 images and 17 drone videos)
Source and characteristics of the images and videos
Data Preprocessing
Annotation process for polygonal labels
Data cleaning and augmentation techniques
Model Selection and Training
Introduction of the YOLOv8-nano model
Training process and hyperparameters
Evaluation metrics used (mAP50, macro average)
Results
Validation Performance
YOLOv8-nano model performance on validation dataset
mAP50 score and its significance
Test Dataset Evaluation
Macro average performance on the test dataset
Comparison with baseline methods
Benchmark Creation
New Benchmark Dataset
Description of the polygonal annotation process
Importance for future research in rip current segmentation
Contribution to the Field
Potential applications and future research directions
Comparison with traditional observation methods
Conclusion
Summary of Findings
Recap of the model's performance and the dataset's utility
Implications
Impact on rip current detection and safety
Scalability and cost-effectiveness of machine learning approaches
Future Work
Suggestions for further research and improvements
Basic info
papers
computer vision and pattern recognition
artificial intelligence
Advanced features
Insights
In what ways do machine learning methods provide advantages over traditional observation methods in rip current detection?
What are the key features of the YOLOv8-nano model that contributed to its performance in rip current instance segmentation?
How does the new benchmark with polygonal annotations enhance the dataset for rip current segmentation?