Substitutional Neural Image Compression
This addresses the challenge of optimizing compression quality and control in neural image compression for applications like media storage and transmission, though it is incremental as it builds on existing neural compression models.
The paper tackles the problem of enhancing neural image compression models by introducing Substitutional Neural Image Compression (SNIC), which replaces the original image with a substitutional one to improve compression performance toward flexible distortion metrics and enable bit-rate control without additional data or tuning, achieving improvements as measured by rate-distortion curves.
We describe Substitutional Neural Image Compression (SNIC), a general approach for enhancing any neural image compression model, that requires no data or additional tuning of the trained model. It boosts compression performance toward a flexible distortion metric and enables bit-rate control using a single model instance. The key idea is to replace the image to be compressed with a substitutional one that outperforms the original one in a desired way. Finding such a substitute is inherently difficult for conventional codecs, yet surprisingly favorable for neural compression models thanks to their fully differentiable structures. With gradients of a particular loss backpropogated to the input, a desired substitute can be efficiently crafted iteratively. We demonstrate the effectiveness of SNIC, when combined with various neural compression models and target metrics, in improving compression quality and performing bit-rate control measured by rate-distortion curves. Empirical results of control precision and generation speed are also discussed.