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

The past few years have witnessed increasing interests in applying deep learning to video compression. However, the existing approaches compress a frame with only number of reference frames, which limits their ability fully exploit temporal correlation among frames. To overcome this shortcoming, paper proposes Recurrent Learned Video Compression (RLVC) approach Auto-Encoder (RAE) and Probability Model (RPM). Specifically, RAE employs recurrent cells both encoder decoder. As such, information large range frames can be used for generating latent representations reconstructing compressed outputs. Furthermore, proposed RPM network recurrently estimates Mass Function (PMF) representation, conditioned on distribution previous representations. Due consecutive conditional cross entropy lower than independent entropy, thus reducing bit-rate. experiments show that our achieves state-of-the-art learned compression performance terms PSNR MS-SSIM. Moreover, outperforms default Low-Delay P (LDP) setting x265 PSNR, also has better MS-SSIM SSIM-tuned slowest x265. codes are available at https://github.com/RenYang-home/RLVC.git.