Attention-YOLOX: Improvement in On-Road Object Detection by Introducing Attention Mechanisms to YOLOX

Deep learning-based Object detection methods have been widely used in autonomous driving in recent years. An object detection algorithm having high detection accuracy and speed is crucial to the safety of autonomous driving vehicles. In this paper, we propose Attention-YOLOX, which embeds an attention mechanism in the object detection algorithm YOLOX to improve the detection accuracy. We propose four models that provide attention mechanisms to capture the inter-channel relationship, the relationship between channel-wise and spatial features, and the long- distance dependency of features, which assist the model to accurately predict the class and location of the target object. We evaluated the effectiveness of the proposed methods on the KITTI, BDDIOOK, and SODAIOM datasets containing images of various scenes and visual conditions. We confirmed that our methods have better detection accuracy than YOLOX with no attention mechanism while maintaining a high processing speed.