Based on Improved U2-Net for Visible to Infrared Image Translation

TWith the rapid development of artificial intelligence technology, significant progress has been made in the field of image generation and translation. However, traditional Generative Adversarial Networks[1] (GAN) face issues such as training instability, mode collapse, and artifacts in image translation tasks. This paper proposes an improved method based on U2-Net[2] for visible to infrared image translation. By introducing the Criss-Cross Attention (CCA) [3]module into the deep layers of U2-Net, the model's ability to capture global information is enhanced, and a frequency domain loss function is used to optimize the image quality. Experimental results show that the improved model outperforms the original U2-Net in multiple evaluation metrics.