An Improved Anchor-Free Object Detection Method
Han Yu, Tonghe Ding, Tianping Li, Meng Li · 2022 International Conference on Machine Learning and Intelligent Systems Engineering (MLISE) · 2022
Object detection plays an important role in various industries. However, a fully convolutional one-stage detector (FCOS) has high computational cost and low detection accuracy. Therefore, in this paper, we fused the attention module (CBAM) with feature pyramid (FPN) to extract important feature information, suppress useless information and improve detection accuracy. Finally, we use inverted residual convolution block to replace the detection head of the original method, and the improved detection head reduces the calculation cost and amount of calculation. We use PASCAL VOC to train and evaluate our network. Experimental results show that, compared with the traditional method, the detection accuracy is improved by 1.2%, and the number of parameters is reduced by 4.2m.