self-supervised image networks

This paper revisits the concept of distilling 2D features into 3D networks for autonomous driving applications. The authors propose a new approach for semantic segmentation that significantly improves upon previous…