Attention-Based Inter-Prediction for Versatile Video Coding

Versatile Video Coding (VVC) is the latest video coding standard, which provides significant efficiency to its successors based on new tools and flexibility. In this paper, we propose a generative adversarial network-based inter-picture prediction approach for VVC. The proposed method involves two major parts, deep attention map estimation frame interpolation. Adjacent VVC-coded frames in every other are taken as reference data prediction. classifies pixels into high-interest low-interest. low-interest replaced by generated from interpolation without extra coded bits, while encoded using conventional VVC tools. generation of interpolated can be incorporated encoding algorithm under unified framework. Experimental results show that improves with moderate increase (26.7%) runtime. An average 1.91% BD-rate savings compared software Random-Access configuration was achieved, where bitrate reduction chroma components (U V) observed.