Comparative Study of RT-DETR Models for Infrared Object Detection: Impact of Data Augmentation

This study evaluates the impact of data augmentation techniques—Mosaic, Mixup, and Copy-paste—on the performance of RT-DETR models for infrared (IR) object detection using the Hanwha IR Thermal Image Dataset. By testing three model configurations, the research highlights how augmentation improves model accuracy and efficiency while addressing IR-specific challenges like background blending and low contrast. Results show that combining Mosaic and Copy-paste enhances generalization, whereas Mixup poses challenges for detecting small objects. The findings provide valuable insights for optimizing data augmentation strategies in real-world IR detection applications.