Real-Time Lightweight CNN for Detecting Road Object of Various Size

This paper proposed a novel lightweight convolutional neural network suitable for road object detection which not only for small objects, but for large objects. The proposed network outperformed detection performance of existing convolutional neural networks on KITTI datasets and satisfied real-time processing speed of 10ms on PC and 65ms on NVIDIA TX2. The model is suitable for running in an embedded environment with only 3-million weight parameters.