End-to-End Learned Image Compression With Augmented Normalizing Flows

Yung-Han Ho, Chih-Chun Chan, Wen-Hsiao Peng, Hsueh-Ming Hang; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 1931-1935

Abstract


This paper presents a new attempt at using augmented normalizing flows (ANF) for lossy image compression. ANF is a specific type of normalizing flow models that augment the input with an independent noise, allowing a smoother transformation from the augmented input space to the latent space. Inspired by the fact that ANF can offer greater expressivity by stacking multiple variational autoencoders (VAE), we generalize the popular VAE-based compression framework by the autoencoding transforms of ANF. When evaluated on Kodak dataset, our ANF-based model provides 3.4% higher BD-rate saving as compared with a VAE-based baseline that implements hyper-prior with mean prediction. Interestingly, it benefits even more from the incorporation of a post-processing network, showing 11.8% rate saving as compared to 6.0% with the baseline plus post-processing.

Related Material


[pdf]
[bibtex]
@InProceedings{Ho_2021_CVPR, author = {Ho, Yung-Han and Chan, Chih-Chun and Peng, Wen-Hsiao and Hang, Hsueh-Ming}, title = {End-to-End Learned Image Compression With Augmented Normalizing Flows}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2021}, pages = {1931-1935} }