Recent work has shown that Variational Autoencoders (VAEs) can be used to upper-bound the information rate-distortion (R-D) function of images, i.e., the fundamental limit of lossy image compression. In this paper, we report an improved upper bound on the R-D function of images implemented by (1) introducing a new VAE model architecture, (2) applying variable-rate compression techniques, and (3) proposing a novel \ourfunction{} to stabilize training. We demonstrate that at least 30\% BD-rate reduction w.r.t. the intra prediction mode in VVC codec is achievable, suggesting that there is still great potential for improving lossy image compression. Code is made publicly available at https://github.com/duanzhiihao/lossy-vae.
@article{arxiv.2309.02574,
title = {An Improved Upper Bound on the Rate-Distortion Function of Images},
author = {Zhihao Duan and Jack Ma and Jiangpeng He and Fengqing Zhu},
journal= {arXiv preprint arXiv:2309.02574},
year = {2023}
}
Comments
Conference paper at ICIP 2023. The first two authors share equal contributions