Abstract

This paper describes a new technique for finding disentangled semantic directions in the latent space of StyleGAN. Our method identifies meaningful orthogonal subspaces that allow editing of one human face attribute, while minimizing undesired changes in other attributes. Our model is capable of editing a single attribute in multiple directions, resulting in a range of possible generated images. We compare our scheme with three state-of-the-art models and show that our method outperforms them in terms of face editing and disentanglement capabilities. Additionally, we suggest quantitative measures for evaluating attribute separation and disentanglement, and exhibit the superiority of our model with respect to those measures.

Method

We use pre-trained attribute classifiers to learn a decomposition of \(\mathcal{W}^+\) latent space into multiple orthogonal subspaces \(\left \{ S_i \right \}_{i=0}^N\), each of which is associated with a single attribute. Thereby, \(\mathcal{W}^+\) is the direct sum of \(S_i\).

Each vector \(\mathbf{w}\) can be uniquely expressed by the sum of its supspace projections.

Multi-directional Edits

Multi-directional editing refers to changing a latent vector within a subspace in different directions. This allows to generate a diverse variety of images that differ in a specific attribute.

Gender

Age

Race

BibTeX


            @article{naveh2023multidirectional,
                  title={Multi-Directional Subspace Editing in Style-Space},
                  author={Chen Naveh and Yacov Hel-Or},
                  year={2023},
                  eprint={2211.11825},
                  archivePrefix={arXiv},
                  primaryClass={cs.CV}
            }

Acknowledgements

This template was originally made by Phillip Isola and Richard Zhang for a colorful ECCV project; the code can be found here.