Virtual Normal: Enforcing Geometric Constraints for Accurate and Robust Depth Prediction

Monocular depth prediction plays a crucial role in understanding 3D scene geometry. Although recent methods have achieved impressive progress in evaluation metrics such as the pixel-wise relative error, most methods neglect the geometric constraints in the 3D space. We show the importance of the high-order 3D geometric constraints for depth prediction. By designing a loss term that enforces a simple geometric constraint, virtual normal directions determined by randomly sampled three points in the reconstructed 3D space, we significantly improve the accuracy and robustness of monocular depth estimation. The virtual normal loss disentangles the scale information and enrich the model with better shape information. We show state-of-the-art results on NYU Depth-V2 and KITTI. Besides, we are now able to recover good 3D structures of the scene such as the point cloud and surface normal directly from the depth with better qualities, eliminating the necessity of training new sub-models as was previously done. Furthermore, when not having absolute metric depth training data, we can use virtual normal to learn a robust affine-invariant depth generated on diverse scenes. We construct a large-scale and diverse dataset for training affine-invariant depth, termed Diverse Scene Depth dataset (DiverseDepth), which consists of a broad range of scenes.