Multilevel Domain Adaptive Detection Method for Haze Scenes
Chonghuan Liu, Ruizhi Liu, Ge Lin, Hailiang Liu, Zhuo Su · 2022
Object detection is a focus in the field of computer vision, but in some complex scenes, such as fog, rain, and night applications, it will still face problems of the domain transfer. Therefore, for unsupervised object detection problems under complex scenes, most of the existing methods are adopted in domain adaptive learning technology to solve. The domain adaptive detection method can avoid spending a lot of money to construct labels for training images with fog. It can also avoid the use of synthetic foggy images but directly use real foggy images for training so as to eliminate the error caused by the synthesis of foggy images. However, the existing domain adaptive detection methods ignore the characteristics of haze scenes. The method presented in this paper will propose a new domain adaptive detection network based on the characteristics of the hazing scene itself and has achieved good results in the experiments.