Multi-feature contrastive learning for unpaired image-to-image translation

Abstract Unpaired image-to-image translation for the generation field has made much progress recently. However, these methods suffer from mode collapse because of overfitting discriminator. To this end, we propose a straightforward method to construct contrastive loss using feature information discriminator output layer, which is named multi-feature learning (MCL). Our proposed enhances performance and solves problem model by further leveraging learning. We perform extensive experiments on several open challenge datasets. achieves state-of-the-art results compared with current methods. Finally, series ablation studies proved that our approach better stability. In addition, also practical single image tasks. Code available at https://github.com/gouayao/MCL.