基于自注意力的无配对图像翻译方法研究

In this paper, a new self-attention-based unpaired image translation method is proposed to solve the generation structure optimization problem existing in current unpaired image translation methods. This method combines multi-head self-attention module and convolutional neural network to enhance the expression ability of the algorithm for global image features. Channel excitation module is used to enhance the ability of multi-head self-attention extracting global channel information. In addition, a new contrast loss constraint is proposed in this paper to ensure the content consistency between the generated image and the original in global and local aspects. Finally, the proposed method is tested on several public data sets, and the experimental results show that the proposed method can effectively improve the image translation results and enhance the authenticity after translation compared with the benchmark method.