CVCP-Fusion: On Implicit Depth Estimation for 3D Bounding Box Prediction