Feature Pyramid Networks for Crowd Counting

摘要

In this paper, we propose the use of Feature Pyramid Networks (FPN) for Crowd Counting problem. FPN previously has been used for retinanet, the state-of-the-art model for object detection. By using FPN, our proposed crowd counting model achieved a state-of-the-art performance for UCF CC 50 dataset with MAE 136.4 and MSE 223.6. The proposed model is also achieved a state-of-the-art MSE value of 7.6 for ShanghaiTech Part B dataset. The code can be accessed at https: //github.com/wawancenggoro/fpncc.