FogGuard is a new object detection network designed to address challenges posed by foggy weather conditions. It utilizes a Teacher-Student Perceptual loss to enhance accuracy in detecting objects in foggy images. The method incorporates synthetic fog generation and fine-tuning on clear and foggy datasets to improve performance. Extensive evaluations demonstrate the superiority of FogGuard over existing methods, achieving higher mean Average Precision (mAP) on datasets like PASCAL VOC and RTTS. The approach ensures robust performance even in adverse weather conditions, without introducing overhead during inference.
A otro idioma
del contenido fuente
arxiv.org
Ideas clave extraídas de
by Soheil Ghara... a las arxiv.org 03-15-2024
https://arxiv.org/pdf/2403.08939.pdfConsultas más profundas