MFRNet: A New CNN Architecture for Post-Processing and In-loop Filtering

In this paper, we propose a novel convolutional neural network (CNN) architecture, MFRNet, for post-processing (PP) and in-loop filtering (ILF) in the context of video compression. This consists four Multi-level Feature review Residual dense Blocks (MFRBs), which are connected using cascading structure. Each MFRB extracts features from multiple layers connections multi-level residual learning order to further improve information flow between these blocks, each them also reuses high dimensional previous MFRB. has been integrated into PP ILF coding modules both HEVC (HM 16.20) VVC (VTM 7.0), fully evaluated under JVET Common Test Conditions Random Access configuration. The experimental results show significant consistent gains over anchor codecs (HEVC HM VTM) other existing CNN-based PP/ILF approaches based on Bjontegaard Delta measurements PSNR VMAF quality assessment. When MFRNet is 16.20, up 16.0% (BD-rate VMAF) demonstrated ILF, 21.0% PP. respective VTM 7.0 5.1% 7.1%