Affine Direct/Skip Mode with Motion Vector Differences in Video Coding

Tianliang Fu, Kai Zhang, Hongbin Liu, Li Zhang, Shanshe Wang, Siwei Ma, Wen Gao

Abstract: Audio and Video coding Standard-3 (AVS-3) is an emerging next-generation video coding standard beyond AVS-2. New technologies, such as affine motion compensation and ultimate motion vector expression have been adopted in AVS-3 phase2. In this paper, Affine Direct/Skip mode with Motion Vector Differences (ADS-MVD) is proposed to improve the existing affine direct/skip mode in AVS-3 phase2. With ADS-MVD, control point motion vectors of a chosen direct/skip merge candidate are refined by adding offsets. The offsets are selected from predefined offset candidates and the indices of the selected offsets are signaled. Simulation results demonstrate that the proposed method can achieve 0.20% BD-rate savings for Y components on average under the random-access configurations with a negligible encoding time/decoding time change. On sequences with rich affine motions, the proposed method achieves a coding gain of up to 0.45%. The method of ADS-MVD has been adopted into AVS-3 phase2.