PTR-CNN for in-loop filtering in video coding

Tong Shao, Tianqi Liu, Dapeng Wu, Chia-Yang Tsai, Zhijun Lei, Ioannis Katsavounidis

Abstract: Highlights•A non-learning image quality assessment method is used to select a reference frame.•The reference frame serves as extra input and guidance to the deep CNN filter.•Transform unit partition and prediction residual are used as extra auxiliary inputs.•The model slightly outperforms MIF-Net in RD performance and has lower complexity.•Our proposed method also improved subjective visual quality.