EYOLOX: An Efficient One-Stage Object Detection Network Based on YOLOX

Object detection has drawn the attention of many researchers due to its wide application in computer vision-related applications. In this paper, a novel model is proposed for object detection. Firstly, new neck designed model, including an efficient SPPNet (Spatial Pyramid Pooling Network), modified NLNet (Non Local Network) and lightweight adaptive feature fusion module. Secondly, head with double residual branch structure presented reduce delay decoupled improve ability. Finally, these improvements are embedded YOLOX as plug-and-play modules forming high-performance detector, EYOLOX (EfficientYOLOX). Extensive experiments demonstrate that achieves significant improvements, which increases YOLOX-s from 40.5% 42.2% AP on MS COCO dataset single GPU. Moreover, performance also outperforms YOLOv6 some SOTA methods same number parameters GFLOPs. particular, only been trained COCO-2017 without using any other datasets, pre-training weights backbone part loaded.