Split Hierarchical Variational Compression
This work addresses image compression challenges for applications requiring efficient and practical coding, though it appears incremental as it builds upon existing VAE and bits-back coding frameworks.
The paper tackled the limitations of variational autoencoders (VAEs) in image compression, such as coding practicalities and compression ratios, by introducing Split Hierarchical Variational Compression (SHVC), which achieved state-of-the-art performance in lossless image compression with up to 100x fewer parameters than competing VAE methods.
Variational autoencoders (VAEs) have witnessed great success in performing the compression of image datasets. This success, made possible by the bits-back coding framework, has produced competitive compression performance across many benchmarks. However, despite this, VAE architectures are currently limited by a combination of coding practicalities and compression ratios. That is, not only do state-of-the-art methods, such as normalizing flows, often demonstrate out-performance, but the initial bits required in coding makes single and parallel image compression challenging. To remedy this, we introduce Split Hierarchical Variational Compression (SHVC). SHVC introduces two novelties. Firstly, we propose an efficient autoregressive prior, the autoregressive sub-pixel convolution, that allows a generalisation between per-pixel autoregressions and fully factorised probability models. Secondly, we define our coding framework, the autoregressive initial bits, that flexibly supports parallel coding and avoids -- for the first time -- many of the practicalities commonly associated with bits-back coding. In our experiments, we demonstrate SHVC is able to achieve state-of-the-art compression performance across full-resolution lossless image compression tasks, with up to 100x fewer model parameters than competing VAE approaches.