Multi-Style Cartoonization: Leveraging Multiple Datasets With GANs

Cai, Jianlu; Li, Frederick W. B.; Nan, Fangzhe; Yang, Bailin

Abstract

Scene cartoonization aims to convert photos into stylized cartoons. While GANs can generate high-quality images, previous methods focus on individual images or single styles, ignoring relationships between datasets. We propose a novel multi-style scene cartoonization GAN that leverages multiple cartoon datasets jointly. Our main technical contribution is a multi-branch style encoder that disentangles representations to model styles as distributions over entire datasets rather than images. Combined with a multi-task discriminator and perceptual losses optimizing across collections, our model achieves state-of-the-art diverse stylization while preserving semantics. Experiments demonstrate that by learning from inter-dataset relationships, our method translates photos into cartoon images with improved realism and abstraction fidelity compared to prior arts, without iterative re-training for new styles.

Citation

Cai, J., Li, F. W. B., Nan, F., & Yang, B. (2024). Multi-Style Cartoonization: Leveraging Multiple Datasets With GANs. Computer Animation and Virtual Worlds, 35(3), Article e2269. https://doi.org/10.1002/cav.2269

• Journal Article Type: Article
• Acceptance Date: May 1, 2024
• Online Publication Date: May 17, 2024
• Publication Date: May 17, 2024
• Deposit Date: Apr 29, 2024
• Publicly Available Date: May 17, 2024
• Journal: Computer Animation and Virtual Worlds
• Print ISSN: 1546-4261
• Electronic ISSN: 1546-427X
• Publisher: Wiley
• Peer Reviewed: Peer Reviewed
• Volume: 35
• Issue: 3
• Article Number: e2269
• DOI: https://doi.org/10.1002/cav.2269
• Public URL: https://durham-repository.worktribe.com/output/2408024
• Related Public URLs: https://casa2024.wtu.edu.cn/
• Additional Information: Conference name: Computer Animation & Social Agents (CASA) 2024