Paper accepted at ECCV 2020
Pleased to announce that the following paper has been accepted to ECCV 2020 as spotlight!
"Guided Deep Decoder: Unsupervised Image Pair Fusion"
Tatsumi Uezato, Danfeng Hong, Naoto Yokoya, Wei He
In this paper, we propose a deep network used as a regularizer for the various image fusion tasks. The proposed network exploits spatial details and semantic features from a guidance image and generates an output guided by the extracted features. Unlike conventional methods, the proposed network enables the network parameters to be optimized in an unsupervised way without training data. Our results show that the proposed network-based regularizer can achieve state-of-the-art performance in various image fusion problems.
Illustration of image pair fusion