Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials

TL;DR

This paper introduces a highly efficient approximate inference algorithm for fully connected CRFs with Gaussian edge potentials, significantly improving image segmentation accuracy while handling billions of pixel connections.

Contribution

It presents a novel, scalable inference method for fully connected CRFs with Gaussian kernels, enabling practical use of dense pixel-level models.

Findings

Dense connectivity improves segmentation accuracy

The algorithm handles billions of edges efficiently

Experimental results show significant accuracy gains

Abstract

Most state-of-the-art techniques for multi-class image segmentation and labeling use conditional random fields defined over pixels or image regions. While region-level models often feature dense pairwise connectivity, pixel-level models are considerably larger and have only permitted sparse graph structures. In this paper, we consider fully connected CRF models defined on the complete set of pixels in an image. The resulting graphs have billions of edges, making traditional inference algorithms impractical. Our main contribution is a highly efficient approximate inference algorithm for fully connected CRF models in which the pairwise edge potentials are defined by a linear combination of Gaussian kernels. Our experiments demonstrate that dense connectivity at the pixel level substantially improves segmentation and labeling accuracy.