SSDFuse: A Semantic Segmentation-Driven Infrared and Visible Image Fusion Method

In the field of computer vision, the fusion of infrared and visible images plays a crucial role. Conventional approaches frequently put visual improvement ahead of the requirements of high-level vision tasks. We present the Semantic Segmentation-Driven Fusion Network (SSDFuse), which uses semantic guidance to bridge the gap between advanced vision tasks and image fusion, in order to address these issues. SSDFuse incorporates a semantic segmentation module to provide global context and detailed local information for the fusion process. Additionally, it employs self-attention mechanisms to effectively integrate information, producing high-quality images beneficial for downstream semantic segmentation tasks. Our training strategy maximizes the utilization of semantic guidance from high-level vision tasks. Extensive experiments demonstrate that SSDFuse achieves promising results in infrared-visible image fusion and enhances the performance of downstream semantic segmentation on a unified benchmark.