Few-Shot Domain Adaptation for Learned Image Compression

TL;DR

This paper introduces a few-shot domain adaptation technique for learned image compression, significantly improving performance on out-of-domain images with minimal target data and fewer parameters.

Contribution

It proposes a novel plug-and-play adapter approach that enhances pre-trained LIC models for domain adaptation using lightweight convolutional and low-rank adapters.

Findings

Achieves performance comparable to H.266/VVC with only 25 target samples.

Matches full-model finetuning performance with less than 2% additional parameters.

Significantly improves out-of-domain image compression quality.

Abstract

Learned image compression (LIC) has achieved state-of-the-art rate-distortion performance, deemed promising for next-generation image compression techniques. However, pre-trained LIC models usually suffer from significant performance degradation when applied to out-of-training-domain images, implying their poor generalization capabilities. To tackle this problem, we propose a few-shot domain adaptation method for LIC by integrating plug-and-play adapters into pre-trained models. Drawing inspiration from the analogy between latent channels and frequency components, we examine domain gaps in LIC and observe that out-of-training-domain images disrupt pre-trained channel-wise decomposition. Consequently, we introduce a method for channel-wise re-allocation using convolution-based adapters and low-rank adapters, which are lightweight and compatible to mainstream LIC schemes. Extensive experiments across multiple domains and multiple representative LIC schemes demonstrate that our method significantly enhances pre-trained models, achieving comparable performance to H.266/VVC intra coding with merely 25 target-domain samples. Additionally, our method matches the performance of full-model finetune while transmitting fewer than $2\%$ of the parameters.