Abstract

Representation disentanglement is an important goal of the representation learning that benefits various of downstream tasks. To achieve this goal, many unsupervised learning representation disentanglement approaches have been developed. However, the training process without utilizing any supervision signal have been proved to be inadequate for disentanglement representation learning. Therefore, we propose a novel weakly-supervised training approach, named as SW-VAE, which incorporates pairs of input observations as supervision signal by using the generative factors of datasets. Furthermore, we introduce strategies to gradually increase the learning difficulty during training to smooth the training process. As shown on several datasets, our model shows significant improvement over state-of-the-art (SOTA) methods on representation disentanglement tasks.

Metadata

publication

European Conference on Computer Vision, 73-87, 2022

year

2022

publication date

2022/10/23

authors

Jiageng Zhu, Hanchen Xie, Wael Abd-Almageed

link

https://link.springer.com/chapter/10.1007/978-3-031-25063-7_5

resource_link

https://arxiv.org/pdf/2209.10623

book

European Conference on Computer Vision

pages

73-87

publisher

Springer Nature Switzerland