Style Transfer of Chinese Traditional Dermatoglyphic Pattern with Wide Diffusion Model

This paper investigates the style transfer of Chinese Traditional Patterns based on diffusion models. Chinese Traditional Patterns boast a rich history and unique artistic style, yet their application and dissemination face challenges. We collected and organized images of Chinese Traditional Patterns, constructing a high-quality Chinese Traditional Dermatoglyphic Pattern Dataset(CTDPD), and proposed the Wide Diffusion Model(WDM) with the Wide AutoEncoder to provide comprehensive receptive field information. The results from the experiment demonstrate that the Wide Diffusion Model achieves outstanding performance on the CTDPD dataset. Our designed WDM achieves the current SOTA performance compared to other cutting-edge methods on CTDPD dataset with Frechet Inception Distance of 5.45 and Inception Score of 207.5. This work thus offers significant backing to the heritability and creation of the Chinese Traditional Patterns.