Pixel-Pair Occlusion Relationship Map(P2ORM): Formulation, Inference & Application

TL;DR

This paper introduces a unified formulation for geometric occlusion in 2D images using pixel-pair relations, enabling accurate occlusion datasets, pixel-level estimation, and improved depth map refinement.

Contribution

It proposes a novel pixel-pair occlusion relation formulation for occlusion boundary and orientation estimation from single images, advancing geometric occlusion understanding.

Findings

Outperforms existing methods in occlusion relationship estimation

Enables large-scale accurate occlusion dataset generation

Improves monocular depth map refinement performance

Abstract

We formalize concepts around geometric occlusion in 2D images (i.e., ignoring semantics), and propose a novel unified formulation of both occlusion boundaries and occlusion orientations via a pixel-pair occlusion relation. The former provides a way to generate large-scale accurate occlusion datasets while, based on the latter, we propose a novel method for task-independent pixel-level occlusion relationship estimation from single images. Experiments on a variety of datasets demonstrate that our method outperforms existing ones on this task. To further illustrate the value of our formulation, we also propose a new depth map refinement method that consistently improve the performance of state-of-the-art monocular depth estimation methods. Our code and data are available at http://imagine.enpc.fr/~qiux/P2ORM/.