VO-Monodepth: Enhancing self-supervised monocular depth estimation with traditional visual odometry

December 2019

tl;dr: Use sparse density measurement from VO algorithm to enhance depth estimation.

Overall impression

This paper combines the idea of depth estimation with depth completion.

Key ideas

The paper used a sparsity invariant autoencoder to densify the sparse measurement before concatenating the sparse data with RGB input.
Inner Loss: between SD (sparse depth) and DD (denser depth after sparse conv)
Outer loss: between SD and d (dense estimation) on where the SD is defined.

Technical details

VO pipeline only provides ~0.06% of sparse depth measurement.
Sparcity invariant CNNs performs weighted average only on valid inputs. This makes the network invariant to input sparsity.

Notes

Best supervised mono depth estimation: DORN
Scale recovery method is needed for monodepth estimation and any mono VO methods.
Both ORB-SLAM v1 and v2 supports mono and stereo.