Multiple Dilation과 Feature Summation을 이용한 객체 검출 백본 네트워크

초록

The advancement of deep learning leads to the trend of using very deep network. But, it is not practical in real project, especially for people who have limited resources or real-time requirement. In this paper, we propose a new backbone network for object detection combined with multiple dilation and feature summation. By using feature summation, we can prevent loss of spatial information that is caused by convolving. And we can widen the receptive field of individual neurons without adding more parameters by using multiple dilated convolution. And by using shallow neural network as backbone network, our network can be trained and used in environment with limited resources and without pretraining it in ImageNet dataset. Based on our experiment, our network got 71% accuracy.