arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2403.18535eess.IVcs.LG
收藏

面向神经图像编解码器的理论界限引导的分层 VAE

Theoretical Bound-Guided Hierarchical VAE for Neural Image Codecs

Elmore Family School of Electrical and Computer Engineering(埃尔莫尔电气与计算机工程学院);Purdue University(普渡大学)

Yichi Zhang, Zhihao Duan, Yuning Huang, Fengqing Zhu

AI总结:

为缩小变分自编码器理论界限与现有神经图像编解码器性能间的差距,提出理论界限引导的分层 VAE(BG-VAE),利用该界限指导模型优化,实现性能优于现有方法的变码率编解码器。

摘要:

近期研究揭示了变分自编码器(VAEs)与率失真理论之间的重要理论联系,特别是在利用 VAEs 估计图像信息率失真函数的理论上限方面。此类估计的理论界限显著优于现有神经图像编解码器(NICs)的性能。为缩小这一差距,我们提出了一种理论界限引导的分层 VAE(BG-VAE)用于 NIC。所提出的 BG-VAE 利用理论界限引导 NIC 模型提升性能。我们使用分层 VAEs 实现了 BG-VAE,并通过大量实验证明了其有效性。结合先进的神经网络模块,我们提供了一种灵活的变码率 NIC,在兼顾率失真性能和计算复杂度时优于现有方法。代码可在 BG-VAE 获取。

英文摘要:

Recent studies reveal a significant theoretical link between variational autoencoders (VAEs) and rate-distortion theory, notably in utilizing VAEs to estimate the theoretical upper bound of the information rate-distortion function of images. Such estimated theoretical bounds substantially exceed the performance of existing neural image codecs (NICs). To narrow this gap, we propose a theoretical bound-guided hierarchical VAE (BG-VAE) for NIC. The proposed BG-VAE leverages the theoretical bound to guide the NIC model towards enhanced performance. We implement the BG-VAE using Hierarchical VAEs and demonstrate its effectiveness through extensive experiments. Along with advanced neural network blocks, we provide a versatile, variable-rate NIC that outperforms existing methods when considering both rate-distortion performance and computational complexity. The code is available at BG-VAE.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑