A progressive CNN in-loop filtering approach for inter frame coding

作者：

Highlights：

• CNN models from direct training are local optimal and effective for intra coding.

• Direct models cause over-filtering in inter coding.

• Develop a progressive training method to produce inter in-loop filtering models.

• Propose a frame-level model selection strategy for the high-bitrate coding scenario.

摘要

•CNN models from direct training are local optimal and effective for intra coding.•Direct models cause over-filtering in inter coding.•Develop a progressive training method to produce inter in-loop filtering models.•Propose a frame-level model selection strategy for the high-bitrate coding scenario.

论文关键词：CNN,In-loop filtering,Model training,Inter coding

论文评审过程：Received 21 August 2020, Revised 2 December 2020, Accepted 7 February 2021, Available online 14 February 2021, Version of Record 18 February 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116201