Broadcast Radar and Camera Fusion Network for 3D Detection Task

In the autonomous driving scenario, millimeter wave radar can provide point clouds data to indicate the position information of objects. But radar point clouds are too sparse and lack of the color and texture information of object. In this paper, we propose a detection method fusing radar and camera data in the feature level. As radar points are sparse, we propose a broadcast mechanism to solve the sparsity of radar data. In addition, we add ASPP in the depth branch and propose an instance aware module in the context branch to make them fit the specific task better. We demonstrate the effectiveness of our method in the experiments section.