Just-in-time reconstruction: Inpainting sparse maps using single view depth predictors as priors (with Saroj Weerasekera, Thanuja Dharmasiri, Ravi Garg and Ian Reid)

This paper presents a method for combining a monocular depth estimate from a CNN with sparse accurate depths sensed from a LIDAR or from key point SLAM.  The method respects the sparse point estimates and the relative depths of the monocular CNN estimate.

[ICRA 2018 paper]