Style-CCL: Content-Preserving Style Transfer via Curriculum Continual Learning
Researchers have developed Style-CCL, a novel framework for content-preserving style transfer using Diffusion Transformers. This method employs a curriculum continual learning approach, training a dual-branch SC-DiT model first on semantic styles and then on texture styles, while also progressing from clean to synthetic data. To mitigate catastrophic forgetting, Random Memory Rehearsal is utilized across training stages. Experiments show Style-CCL outperforms existing methods in style similarity, content consistency, and aesthetic quality. AI
IMPACT Introduces a novel curriculum learning approach for Diffusion Transformers, potentially improving image generation and style transfer capabilities.