TSF: Two-Stage Sequential Fusion for 3D Object Detection

There have been significant advances in 3D object detection using LiDAR and camera fusion for autonomous driving. However, it is surprisingly difficult to effectively design location strategies point cloud-based networks. In this paper, we propose a novel two-stage sequential (TSF) method. the first stage of fusion, TSF generates enhanced cloud by combining raw semantic information image instance segmentation. second stage, proposals generated baseline used complete No-Maximum Suppression (NMS) together with 2D results. Numerous experiments on KITTI validation set show that our method outperforms state-of-the-art multimodal fusion-based methods three classes performance (Easy, Moderate, Hard): cars (89.94%, 82.76%, 76.04), pedestrians (70.74%, 63.47%, 56.56%), cyclists (84.72%, 64.22%, 56.78%). ablation, analyze augmented effect module capability, study best trade-off between running time accuracy.