A Video Coding Solution with Neural-network Enhancement for CLIC 2025

Keji Chen, Licheng Ma, Haixin Wang, Jing Chen, Gang Li, Xiuxin Dou, Linyan Jiang, Xianguo Zhang, Yaqing Li

Abstract: In this work, a video coding solution with Neural-Network (NN) enhancement is proposed for the Challenge on Learned Image Compression (CLIC) 2025. The proposed solution is made up of ESRGAN-based video enhancement pre-processing and improved Enhanced Compression Model (ECM) codec, and the team name of the solution is $\mathit{TCM}$. The improvements of ECM include advanced Rate Control (RC) methods, frame parallel processing, and speed optimization of NN based in-loop Filter (NNLF). According to the pre-evaluation results, the proposed methods achieve more than 0.53 Mean Opinion Score (MOS) improvement compared with VTM baseline.

Team Name: TCM

Submission Number: 6