Inspired by both MVX-Net and PointPillars, image features from a 2D CNN-based feature map are fused with the 3D point cloud in an early fusion architecture. By changing the heavy 3D convolutions of ...