Fig. 1: The framework of CAUGAN.

Note: Fig. 1 shows the architecture of the CAUGAN method for mural image inpainting. It consists of two main parts: the generator and the discriminator. The generator is responsible for inpainting the image. It includes convolution layers and Coordinated Attention Aggregate Transformation (CAAT) blocks. The CAAT blocks allow the generator to capture contextual information from distant areas of the image, helping it generate more realistic inpainting. The input to the generator is damaged and masked image, and the output is the inpainted image. The discriminator is based on the U-Net architecture and serves to evaluate the generated inpainted image. The discriminator compares the inpainted image to the original image (ground truth) and decides whether the generated content is “Real or Fake.” This adversarial process encourages the generator to improve its inpainting quality. Together, these components work to significantly enhance the visual quality and realism of the inpainted images by refining both the generated content and the evaluation process.