MUST-GAN: Multi-level Statistics Transfer for Self-driven Person Image Generation
This work simplifies data preparation for pose-guided person image generation, benefiting researchers and developers by reducing the need for labor-intensive paired datasets.
This paper addresses the challenge of pose-guided person image generation without requiring paired source-target images for supervision. The authors propose a multi-level statistics transfer model that disentangles and transfers appearance features, merging them with pose features to reconstruct source images, thereby enabling self-driven generation. Experiments on the DeepFashion dataset show their method outperforms state-of-the-art supervised and unsupervised methods.
Pose-guided person image generation usually involves using paired source-target images to supervise the training, which significantly increases the data preparation effort and limits the application of the models. To deal with this problem, we propose a novel multi-level statistics transfer model, which disentangles and transfers multi-level appearance features from person images and merges them with pose features to reconstruct the source person images themselves. So that the source images can be used as supervision for self-driven person image generation. Specifically, our model extracts multi-level features from the appearance encoder and learns the optimal appearance representation through attention mechanism and attributes statistics. Then we transfer them to a pose-guided generator for re-fusion of appearance and pose. Our approach allows for flexible manipulation of person appearance and pose properties to perform pose transfer and clothes style transfer tasks. Experimental results on the DeepFashion dataset demonstrate our method's superiority compared with state-of-the-art supervised and unsupervised methods. In addition, our approach also performs well in the wild.