A Feature Adversarial Domain Adaptation Method in Single-Stage Object Detection Models
YU Han-ying, Yi Su, Han Shijiao · 2024
Thanks to the rapid progress of automatic driving technology, the problem of object detection in automatic driving scenarios has become one of the hottest research topics at present. Automatic driving scenarios face different day and night scenes, complex and variable weather conditions, which further increase the demand for the generalization ability of detection models. At the same time, the automatic driving scenario requires a high response speed for the model, and the object detection model in this scenario is usually a single-stage model. In this paper, a feature adversarial domain adaptation method was proposed for single-stage models, and experiments are conducted on multiple datasets to verify the effectiveness of the proposed algorithm.