Causal Contextual Prediction for Learned Image Compression
Introduction
Over the past several years, we have witnessed the impressive progress of learned image compression. Recent learned image codecs are commonly based on auto-encoders, that first encode an image into low-dimensional latent representations and then decode them for reconstruction. To capture spatial dependencies in the latent space, prior works exploit hyperprior and spatial context model to build an entropy model, which estimates the bit-rate for end-to-end rate-distortion optimization. However, such entropy model is suboptimal from two aspects: (1) It fails to capture global-scope spatial correlations among the latents. (2) Cross-channel relationships of latents are still underexplored.In this paper, we propose a \textit{causal context model} that makes use of cross-channel relationships to generate more informative context and thus enhances the entropy model. Furthermore, we propose a \textit{causal global prediction model}, which is able to find global reference points for accurate prediction of current decoding point. Both our proposed \textit{causal context model} and \textit{causal global prediction model} leverage a causal decoding process to facilitate entropy estimation without requiring transmission of overhead. Besides, we further adopt separate attention module to build more powerful transform networks. Experimental results demonstrate that our full image compression model outperforms standard VVC/H.266 codec on Kodak dataset in terms of both PSNR and MS-SSIM, yielding the state-of-the-art rate-distortion performance.
Download
To get the source code, Please check this website.