eduzhai > Applied Sciences > Engineering >

Learning for Video Compression with Recurrent Auto-Encoder and Recurrent Probability Model

  • Save

... pages left unread,continue reading

Document pages: 17 pages

Abstract: The past few years have witnessed increasing interests in applying deeplearning to video compression. However, the existing approaches compress avideo frame with only a few number of reference frames, which limits theirability to fully exploit the temporal correlation among video frames. Toovercome this shortcoming, this paper proposes a Recurrent Learned VideoCompression (RLVC) approach with the Recurrent Auto-Encoder (RAE) and RecurrentProbability Model (RPM). Specifically, the RAE employs recurrent cells in boththe encoder and decoder. As such, the temporal information in a large range offrames can be used for generating latent representations and reconstructingcompressed outputs. Furthermore, the proposed RPM network recurrently estimatesthe Probability Mass Function (PMF) of the latent representation, conditionedon the distribution of previous latent representations. Due to the correlationamong consecutive frames, the conditional cross entropy can be lower than theindependent cross entropy, thus reducing the bit-rate. The experiments showthat our approach achieves the state-of-the-art learned video compressionperformance in terms of both PSNR and MS-SSIM. Moreover, our approachoutperforms the default Low-Delay P (LDP) setting of x265 on PSNR, and also hasbetter performance on MS-SSIM than the SSIM-tuned x265 and the slowest settingof x265. The codes are available at this https URL.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...