Multiscale Augmented Normalizing Flows for Image Compression
This work addresses image compression for applications requiring high fidelity, though it is incremental as it builds on existing invertible models and hierarchical approaches.
The paper tackled the problem of inefficiency in learning-based image compression at high quality by introducing a hierarchical latent space for augmented normalizing flows, achieving average rate savings of over 7% compared to single-scale models.
Most learning-based image compression methods lack efficiency for high image quality due to their non-invertible design. The decoding function of the frequently applied compressive autoencoder architecture is only an approximated inverse of the encoding transform. This issue can be resolved by using invertible latent variable models, which allow a perfect reconstruction if no quantization is performed. Furthermore, many traditional image and video coders apply dynamic block partitioning to vary the compression of certain image regions depending on their content. Inspired by this approach, hierarchical latent spaces have been applied to learning-based compression networks. In this paper, we present a novel concept, which adapts the hierarchical latent space for augmented normalizing flows, an invertible latent variable model. Our best performing model achieved average rate savings of more than 7% over comparable single-scale models.