HiLLoC: Lossless Image Compression with Hierarchical Latent Variable Models
This work addresses the challenge of efficient lossless compression for large-scale image data, with incremental improvements in applying existing methods to new scales.
The authors tackled the problem of scaling lossless image compression to large photographs by leveraging the generalization of fully convolutional VAE models from 32x32 to larger sizes, achieving state-of-the-art compression on full-size ImageNet images.
We make the following striking observation: fully convolutional VAE models trained on 32x32 ImageNet can generalize well, not just to 64x64 but also to far larger photographs, with no changes to the model. We use this property, applying fully convolutional models to lossless compression, demonstrating a method to scale the VAE-based 'Bits-Back with ANS' algorithm for lossless compression to large color photographs, and achieving state of the art for compression of full size ImageNet images. We release Craystack, an open source library for convenient prototyping of lossless compression using probabilistic models, along with full implementations of all of our compression results.