Paper: UC-Net: Uncertainty Inspired RGB-D Saliency Detection via Conditional Variational Autoencoders
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.
Figure 8. Comparisons of saliency maps. “MH1” and “MH2” are two predictions from M-head. “DP1” and “DP2” are predictions of two random MC-dropout during test. “Ours(1)” and “Ours(2)” are two predictions sampled from our CVAE based model. Different from M-head and MC-dropout, which produce consistent predictions for ambiguous images (5th row), UC-Net can produce diverse predictions.
Jing Zhang, Deng-Ping Fan, Yuchao Dai, Saeed Anwar, Fatemeh Sadat Saleh, Tong Zhang, Nick Barnes. UC-Net: Uncertainty Inspired RGB-D Saliency Detection via Conditional Variational Autoencoders. IEEE Computer Vision and Pattern Recognition (CVPR), Oral, 2020.
Download the full paper here.
For more information, contact us.