UC-Net: Uncertainty Inspired RGB-D Saliency Detection via Conditional Variational Autoencoders

TL;DR

This paper introduces UC-Net, a probabilistic RGB-D saliency detection framework that models annotation uncertainty using conditional variational autoencoders, producing multiple saliency maps and achieving state-of-the-art results.

Contribution

The novel use of conditional variational autoencoders to incorporate uncertainty in RGB-D saliency detection, enabling multiple saliency predictions and improved accuracy.

Findings

Outperforms 18 competing algorithms on six datasets

Generates multiple saliency maps per image

Achieves state-of-the-art performance

Abstract

In this paper, we propose the first framework (UCNet) to employ uncertainty for RGB-D saliency detection by learning from the data labeling process. Existing RGB-D saliency detection methods treat the saliency detection task as a point estimation problem, and produce a single saliency map following a deterministic learning pipeline. Inspired by the saliency data labeling process, we propose probabilistic RGB-D saliency detection network via conditional variational autoencoders to model human annotation uncertainty and generate multiple saliency maps for each input image by sampling in the latent space. With the proposed saliency consensus process, we are able to generate an accurate saliency map based on these multiple predictions. Quantitative and qualitative evaluations on six challenging benchmark datasets against 18 competing algorithms demonstrate the effectiveness of our approach in learning the distribution of saliency maps, leading to a new state-of-the-art in RGB-D saliency detection.