Arbitrary Style Transfer Method With Attentional Feature Distribution Matching
Bin Ge, Zhen Shan Hu, Chen Xia, Junming Guan · Research Square · 2023
Abstract Most arbitrary style transfer methods only consider transferring of the features of the style image and the content image. Although the pixel-wise style transfer is achieved. It is limited to preserve the content structure, the model tends to transfer the style features, the loss of image’s information occur during transfer process. The model incline to transfer the style features and preservation of the content structure is weak. The generated pictures will produce artifacts and patterns of style pictures. In this paper, an attention feature distribution matching method for arbitrary style transfer is proposed. In network architecture, a combination of self-attention mechanism and second-order statistics is used to perform style transfer and the style strengthen block (SSB) enhances the style features of generated images. In loss function, traditional content loss is not used, we integrate the attention mechanism and feature distribution matching to construct loss function. The constraints are strengthened to avoid artifacts in generated image. Qualitative and quantitative experiments demonstrate effectiveness of our method compared with state-of-the-art arbitrary style transfer in improving arbitrary style transfer quality.