SRAGAN: Saliency Regularized and Attended Generative Adversarial Network for Chinese Ink-wash Painting Style Transfer

TL;DR

SRAGAN tackles unpaired Chinese ink-wash painting style transfer by integrating image saliency into a bidirectional GAN framework. It introduces three key components: a saliency IOU loss $L_{SIOU}$ to regularize object content, a saliency regularized generator via SANorm to preserve structure, and a saliency attended discriminator to focus stylization on salient regions. Empirical results show improved FID and Saliency MIOU over NST, GAN-based baselines, and diffusion methods across landscape, horse, and bird tasks, with ablations confirming the value of each saliency component. This work demonstrates that saliency-guided content regularization can significantly enhance both content preservation and style fidelity in artistic I2I translation, suggesting broader applicability to content-sensitive style transfer problems.

Abstract

Recent style transfer problems are still largely dominated by Generative Adversarial Network (GAN) from the perspective of cross-domain image-to-image (I2I) translation, where the pivotal issue is to learn and transfer target-domain style patterns onto source-domain content images. This paper handles the problem of translating real pictures into traditional Chinese ink-wash paintings, i.e., Chinese ink-wash painting style transfer. Though a wide range of I2I models tackle this problem, a notable challenge is that the content details of the source image could be easily erased or corrupted due to the transfer of ink-wash style elements. To remedy this issue, we propose to incorporate saliency detection into the unpaired I2I framework to regularize image content, where the detected saliency map is utilized from two aspects: (\romannumeral1) we propose saliency IOU (SIOU) loss to explicitly regularize object content structure by enforcing saliency consistency before and after image stylization; (\romannumeral2) we propose saliency adaptive normalization (SANorm) which implicitly enhances object structure integrity of the generated paintings by dynamically injecting image saliency information into the generator to guide stylization process. Besides, we also propose saliency attended discriminator which harnesses image saliency information to focus generative adversarial attention onto the drawn objects, contributing to generating more vivid and delicate brush strokes and ink-wash textures. Extensive qualitative and quantitative experiments demonstrate superiority of our approach over related advanced image stylization methods in both GAN and diffusion model paradigms.

Figures (15)

• Figure 1: Comparison between Western oil paintings and traditional Chinese ink-wash paintings.
• Figure 2: Top: real-world images and Chinese ink-wash paintings. Bottom: saliency maps detected by CSNet bib30. The saliency detection deep model pre-trained on real-world images can also segment out salient objects well for Chinese ink-wash paintings.
• Figure 3: Overall architecture of our SRAGAN model.
• Figure 4: Architecture details of our proposed saliency regularized generator.
• Figure 5: Architecture details of our proposed saliency attended discriminator.