Relative Depth Estimation as a Ranking Problem

TL;DR

This paper reformulates the relative depth estimation from a single image as a ranking problem, leveraging ranking literature and introducing a new metric for improved accuracy.

Contribution

It introduces a listwise ranking loss and a new metric for relative depth estimation, enhancing performance by applying ranking techniques.

Findings

Improved depth ranking accuracy over previous methods

Effective application of listwise ranking loss to depth estimation

Introduction of a new metric for pixel depth ranking

Abstract

We present a formulation of the relative depth estimation from a single image problem, as a ranking problem. By reformulating the problem this way, we were able to utilize literature on the ranking problem, and apply the existing knowledge to achieve better results. To this end, we have introduced a listwise ranking loss borrowed from ranking literature, weighted ListMLE, to the relative depth estimation problem. We have also brought a new metric which considers pixel depth ranking accuracy, on which our method is stronger.