Improving The Reconstruction Quality by Overfitted Decoder Bias in Neural Image Compression

TL;DR

This paper introduces a method to enhance neural image compression by fine-tuning the decoder's bias for each image, leading to better reconstruction quality with minimal additional cost.

Contribution

It proposes an instance-based fine-tuning approach for the decoder's bias, improving compression performance without retraining the entire model.

Findings

Achieves 3-5% BD-rate improvement over state-of-the-art methods.

Applicable to any end-to-end neural image compression technique.

Provides better reconstruction quality with slight encoding time increase.

Abstract

End-to-end trainable models have reached the performance of traditional handcrafted compression techniques on videos and images. Since the parameters of these models are learned over large training sets, they are not optimal for any given image to be compressed. In this paper, we propose an instance-based fine-tuning of a subset of decoder's bias to improve the reconstruction quality in exchange for extra encoding time and minor additional signaling cost. The proposed method is applicable to any end-to-end compression methods, improving the state-of-the-art neural image compression BD-rate by $3-5\%$.