Reference Frame Generation Algorithm using Dynamical Learning PredNet for VVC

The reference frame plays an important role in the coding efficiency of inter-frame prediction in versatile video coding (VVC). In particular, the prediction of non-linear motion is a challenge for the motion compensation. In this paper, we proposed a new reference frame generation algorithm using dynamical learning PredNet for the inter prediction. The proposed PredNet which include training and inference is implemented to frame buffer in the VVC encoder. The proposed VVC encoder is achieved by the pipeline processing. The simulation results show that the proposed algorithm can achieve an improvement of 0.58% BD-rate comparing to the original VVC algorithm.