Extended support for subblock-based affine motion compensation of VVC

Subblock-based affine motion model and prediction refinement with optical flow for affine (PROF) enhances support for more complex motions in Versatile Video Coding (VVC). Current syntax for affine motion model supports signaling large rotation (e.g. close to 90/180/270 degree) and flipping, however, engine of affine motion compensation and PROF does not work optimally for these cases. In case of larger rotation, the predicted subblock is not useful and also there is a big discontinuity in the borders of predicted subblock. Furthermore, PROF increases discontinuity. In this study we propose a solution to detect large rotation, harmonize PROF process, and apply post processing to subblock prediction samples. In video-based point cloud coding (V-PCC) use case, where YUV sequences are generated based on point cloud data, -2.81% and -3.12% overall Bjøntegaard Delta bitrate gain is achieved for random-access and low-delay configurations, respectively. However, the method does not have noticeable effect on encoding and decoding time.