Aug, 2023

策略性猎物使敏锐的捕食者:通过生成伪装对象增强伪装物体检测器

TL;DRCamouflaged object detection faces challenges, and this research proposes an adversarial training framework, Camouflageator, to generate more camouflaged objects and improve the detection. They also introduce ICEG, a novel COD method that utilizes internal coherence and edge guidance for better segmentation results. Experimental results show that ICEG outperforms existing detectors, and Camouflageator is flexible in improving various COD detectors, achieving state-of-the-art COD performance.