Evaluating RF-DETR and YOLOv26 for Rip Current Detection and Segmentation

Abstract

This application study evaluates two existing instance-segmentation frameworks, RF-DETR (Region Focused Detection Transformer) and YOLOv26, for rip current detection in complex marine imagery. Both models were fine-tuned from pretrained weights on the same 18,389 training images. YOLOv26 used the 4349 labeled images as its validation/evaluation split, whereas the RF-DETR training workflow divided these images into 2184 validation and 2165 internal-test images; the two RF-DETR subsets were combined for the final evaluation so that both models were assessed on the same 4349 labeled images. No architectural modification to RF-DETR or YOLOv26 is claimed. The study instead examines their observed segmentation accuracy, training behavior, and frame-wise inference speed on RipVIS and two unmanned aerial vehicle (UAV) videos. Under the configurations used, RF-DETR produced higher mask mean average precision (mAP). values, whereas YOLOv26 processed the two videos faster. Because the training schedules differ and the study does not include repeated trials, component ablations, embedded-platform tests, or broad environmental coverage, the results should be interpreted as an exploratory application comparison rather than evidence of a new algorithm or deployment readiness.

Received: 15 May 2026 | Revised: 13 August 2026 | Accepted: 23 August 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support this study are openly available in Hugging Face at https://huggingface.co/datasets/Irikos/RipVIS, Video 1 at https://drive.google.com/file/d/1fp1dJVoWGS9bGr2ZRbRksCHcfGEeSP__/view?usp=drive_link, Video 2 at https://drive.google.com/file/d/15p2tU8HYv2lPau6uHFQ-RbXl0-R4C-sU/view?usp=drive_link, and Output video at https://drive.google.com/drive/folders/1IfpnkifIjXQ4B7n4gnkOR0P_O4AKjpa6?usp=drive_link.

Author Contribution Statement

Van Lam Ho: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Supervision, Project administration. Van Khang Le: Methodology, Software, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization. Xuan Vinh Le: Supervision. Trong Doi Nguyen: Validation, Investigation, Resources, Data curation, Project administration. Trang-Thi Ho: Validation, Supervision.