What You See Is What You Detect: Towards better Object Densification in 3D detection