Recently, unsupervised image-to-image translation methods based on contrastive learning have achieved state-of-the-art results in many tasks. However, in the previous work, the negatives are sampled from the input image itself, which inspires us to design a data augmentation method to improve the quality of the selected negatives. Moreover, retaining the content similarity via patch-wise contrastive learning in the embedding space, the previous methods ignore the domain consistency between the generated image and the real images of target domain. In this paper, we propose a novel unsupervised image-to-image translation framework based on multi-crop contrastive learning and domain consistency, called MCDUT. Specifically, we obtain the multi-crop views via the center-crop and the random-crop to generate the negatives, which can increase the quality of the negatives. To constrain the embeddings in the deep feature space, we formulate a new domain consistency loss, which encourages the generated images to be close to the real images in the embedding space of same domain. Furthermore, we present a dual coordinate attention network by embedding positional information into channel attention, which called DCA. We employ the DCA network in the design of generator, which makes the generator capture the horizontal and vertical global information of dependency. In many image-to-image translation tasks, our method achieves state-of-the-art results, and the advantages of our method have been proven through extensive comparison experiments and ablation research.
翻译:暂无翻译