Temporal context video compression with flow-guided feature prediction

Yiming Wang, Qian Huang, Bin Tang, Huashan Sun, Xiaotong Guo, Zhuang Miao

Abstract: Highlights•Flow-guided feature prediction guides the offset maps learning explicitly.•Temporal context compression uses conditional coding for better performance.•Three-stage training strategy avoids error accumulation.•Our proposed method outperforms the H.266(VTM) in MS-SSIM metric.