Research of Image Arbitrary Style Transfer based on Contrastive Learning
Gao Jingxin, Xiao Luo, Han RunPing · 2024
With the practical applications of image style transfer techniques in fields such as image editing and creative design, arbitrary style transfer has become an increasingly hot research area in computer vision. Addressing the challenges in existing arbitrary style transfer models, namely how to maintain consistency in preserving the semantic content and structural information of images during style transfer, while transferring more style features, this study incorporates the idea of contrastive learning into the model training process. By continuously optimizing the content similarity of generated images through contrastive learning loss, the effectiveness of arbitrary style transfer is ensured. Experimental results demonstrate that the generated images of the arbitrary style transfer model designed in this study exhibit superior performance in both preserving content structures and transferring style features compared to existing algorithmic models.