Few-Shot Domain Adaptation for Learned Image Compression

Tianyu Zhang, Haotian Zhang, Yuqi Li, Li Li, Dong Liu
{"title":"Few-Shot Domain Adaptation for Learned Image Compression","authors":"Tianyu Zhang, Haotian Zhang, Yuqi Li, Li Li, Dong Liu","doi":"arxiv-2409.11111","DOIUrl":null,"url":null,"abstract":"Learned image compression (LIC) has achieved state-of-the-art rate-distortion\nperformance, deemed promising for next-generation image compression techniques.\nHowever, pre-trained LIC models usually suffer from significant performance\ndegradation when applied to out-of-training-domain images, implying their poor\ngeneralization capabilities. To tackle this problem, we propose a few-shot\ndomain adaptation method for LIC by integrating plug-and-play adapters into\npre-trained models. Drawing inspiration from the analogy between latent\nchannels and frequency components, we examine domain gaps in LIC and observe\nthat out-of-training-domain images disrupt pre-trained channel-wise\ndecomposition. Consequently, we introduce a method for channel-wise\nre-allocation using convolution-based adapters and low-rank adapters, which are\nlightweight and compatible to mainstream LIC schemes. Extensive experiments\nacross multiple domains and multiple representative LIC schemes demonstrate\nthat our method significantly enhances pre-trained models, achieving comparable\nperformance to H.266/VVC intra coding with merely 25 target-domain samples.\nAdditionally, our method matches the performance of full-model finetune while\ntransmitting fewer than $2\\%$ of the parameters.","PeriodicalId":501289,"journal":{"name":"arXiv - EE - Image and Video Processing","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - EE - Image and Video Processing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.11111","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

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.
用于学习型图像压缩的少镜头域自适应技术
学习图像压缩(LIC)已经实现了最先进的速率-失真性能,被认为有望成为下一代图像压缩技术。然而,预训练的 LIC 模型在应用于训练域外图像时通常会出现明显的性能下降,这意味着它们的泛化能力较差。为了解决这个问题,我们提出了一种通过将即插即用适配器集成到预先训练的模型中来实现 LIC 的少镜头域适应方法。从潜在信道和频率成分之间的类比中汲取灵感,我们对 LIC 中的域间隙进行了研究,发现训练域外的图像会破坏预先训练的信道化分解。因此,我们介绍了一种使用基于卷积的适配器和低阶适配器进行信道明智分配的方法,这种方法重量轻,与主流 LIC 方案兼容。跨越多个域和多个代表性 LIC 方案的广泛实验表明,我们的方法显著增强了预训练模型,仅用 25 个目标域样本就实现了与 H.266/VVC 内编码相当的性能。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信