Rethinking Content and Style: Exploring Bias for Unsupervised Disentanglement

Xuanchi Ren, Tao Yang, Wenjun Zeng, Yuwang Wang

Keywords: Unsupervised Disentanglement, Content and Style Disentanglement, Inductive Bias, Representation Learning

Abstract: Content and style (C-S) disentanglement intends to decompose the underlying explanatory factors of objects into two independent latent spaces. Aiming for unsupervised disentanglement, we introduce an inductive bias to our formulation by assigning different and independent roles to content and style when approximating the real data distributions. The content embeddings of individual images are forced to share a common distribution. The style embeddings encoding instance-specific features are used to customize the shared distribution. The experiments on several popular datasets demonstrate that our method achieves the state-of-the-art disentanglement compared to other unsupervised approaches and comparable or even better results than supervised methods. Furthermore, as a new application of C-S disentanglement, we propose to generate multi-view images from a single view image for 3D reconstruction.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

One-sentence Summary: Aiming for unsupervised disentanglement, we introduce an inductive bias by assigning different and independent roles to content and style when approximating the real data distributions.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2102.10544/code)

Reviewed Version (pdf): https://openreview.net/references/pdf?id=HDLLhR3cFw