RepVGGFuse: an approach for infrared and visible image fusion network based on RepVGG architecture

In this paper, we propose an infrared and visible image fusion network based on RepVGG architecture. This network adopts an encoder-decoder structure. The encoding network, which contains five RepVGG blocks, is utilized to extract deep features of infrared and visible images. Each layer of RepVGG blocks is constructed with 3x3, 1x1 and identity branches while training and converted to single-branch architecture constructed with 3x3 convolutional layers while inferring. These extracted features are added and the fusion image is reconstructed by the decoding network. The proposed method was compared with seven fusion methods and the result shows that the proposed fusion method can retain more contour and texture information with less noise. The proposed method is superior to the comparison methods. The code of the proposed fusion network is available at https://github.com/xiongzhangzzz/repvggfuse.