Fig. 4: Example of DunHuang-Mural dataset.

a DunHuang-Mural images, b Maked Image. Note: The figure shows an example from the DunHuang-Mural dataset. The Dunhuang-Mural dataset was developed from a publicly available Kaggle dataset, accessible via the following link: https://www.kaggle.com/jacobok/datasets and https://www.kaggle.com/datasets/xuhangc/dunhuang-grottoes-painting-dataset-and-benchmark. Through systematic preprocessing using Python, 7983 high-resolution images were generated. This dataset, referred to as the Dunhuang-Mural dataset, provides a critical resource for tasks such as mural image inpainting, damage detection, and validation in the field of digital heritage preservation. In (a) DunHuang-Mural images we have the original DunHuang-Mural images. These are pieces of artwork that include intricate details and colors, showcasing the rich cultural heritage of the Dunhuang murals. The images depict scenes with various figures and elements, such as a seated figure and architectural elements. In (b) Maked Image, the irregular masked images are shown. These masks are applied to the original images, simulating regions that are missing or damaged. The white areas in the mask represent the regions that are missing content and need to be inpainted or restored. The mask highlights the gaps in the images that the inpainting method, such as CAUGAN, will focus on filling to generate a complete and realistic image.