Efficient Nonlinear Transforms for Lossy Image Compression

TL;DR

This paper evaluates Sadam and GDN techniques in nonlinear neural network-based image compression, demonstrating their roles in stabilizing training and enhancing transform capacity, with open-source implementation provided.

Contribution

It provides the first comparative assessment of Sadam and GDN, highlighting their effectiveness in stabilizing training and improving nonlinear transform coding for image compression.

Findings

Sadam and GDN improve training stability of neural transforms.

Both techniques enhance the approximation of rate-distortion optimal transforms.

Open-source code is provided for reproducibility.

Abstract

We assess the performance of two techniques in the context of nonlinear transform coding with artificial neural networks, Sadam and GDN. Both techniques have been successfully used in state-of-the-art image compression methods, but their performance has not been individually assessed to this point. Together, the techniques stabilize the training procedure of nonlinear image transforms and increase their capacity to approximate the (unknown) rate-distortion optimal transform functions. Besides comparing their performance to established alternatives, we detail the implementation of both methods and provide open-source code along with the paper.