DLMFCOS: Efficient Dual-Path Lightweight Module for Fully Convolutional Object Detection

Recent advances in convolutional neural network (CNN)-based object detection have a trade-off between accuracy and computational cost various industrial tasks essential consideration. However, the fully one-stage detector (FCOS) demonstrates low compared with its costs owing to loss of low-level information. Therefore, we propose module called dual-path lightweight (DLM) that efficiently utilizes In addition, DLMFCOS based on DLM achieve an optimal accuracy. Our minimizes feature by extracting spatial channel information parallel implementing bottom-up pyramid improves detection. Additionally, structure head is improved minimize cost. The proposed method was trained evaluated fine-tuning parameters through experiments using public datasets PASCAL VOC 07 MS COCO 2017 datasets. average precision (AP) metric used for our quantitative evaluation matrix performance, model achieves 1.5% improvement at about 33.85% lower each dataset than conventional method. Finally, efficiency verified comparing ablation study.