Learning Controllable Disentangled Representations with Decorrelation Regularization
This addresses the challenge of controllable image editing for applications like face manipulation, though it appears incremental.
The paper tackles the problem of controlling disentanglement in image representations for editing while preserving object identity, and demonstrates that their model effectively disentangles factors of variation and manipulates face images to synthesize desired attributes.
A crucial problem in learning disentangled image representations is controlling the degree of disentanglement during image editing, while preserving the identity of objects. In this work, we propose a simple yet effective model with the encoder-decoder architecture to address this challenge. To encourage disentanglement, we devise a distance covariance based decorrelation regularization. Further, for the reconstruction step, our model leverages a soft target representation combined with the latent image code. By exploiting the real-valued space of the soft target representations, we are able to synthesize novel images with the designated properties. We also design a classification based protocol to quantitatively evaluate the disentanglement strength of our model. Experimental results show that the proposed model competently disentangles factors of variation, and is able to manipulate face images to synthesize the desired attributes.