GMamba: A Generative Adversarial Network with State Space Model for Infrared and Visible Image Fusion

Visible images primarily reflect the color and surface details of objects, while infrared images carry thermal radiation information. Combining these two types of information into one image is advantageous, as it not only provides good thermal radiation information but also includes abundant texture details. The recent development of state space models has thoroughly transformed the field of computer vision. Building upon this foundation, we introduce GMamba, a state space model utilizing Generative Adversarial Networks (GAN) for the fusion of infrared and visible images. The purpose of us is designed to efficiently capture details regarding thermal radiation from infrared images while retaining data concerning texture from visible images. The generator encompasses modules of extraction of features and feature aggregation recasting, making use of the Visual State Space Model (VMamba) to harvest features from both infrared and visible spectrum, ensuring feature alignment during the fusion process. We have also implemented a combination of layer normalization and convolution approaches to boost the model's generalization competence, resulting in the generation of higher-quality fused images. The discriminator part adopts dual discriminators. Through extensive experiments on different datasets, our proposed method surpasses many advanced techniques and achieves comparably good fusion results.