Motion vector refinement for temporal motion vector prediction in video coding

Abstract

Techniques for encoding or decoding video data related to motion vector prediction are described. A video coder determines a collocated picture, a collocated block, and a current reference picture for a current block. The video coder determines a scaled motion vector indicating a position in the current reference picture. The video coder searches a search range to identify a block of the current reference picture associated with a minimum difference with the collocated block. A temporal motion vector predictor (TMVP) motion vector is determined based on scaling the refined motion vector. The video coder derives a predictor for the current block using the TMVP motion vector and encodes or decodes the current block using the derived predictor.

Images