Motion vector prediction using co-located prediction units

A method comprising:

• accessing a target reference picture available for a current pixel group in a current picture;
• determining that a co-located pixel group of the target reference picture has an assigned candidate motion vector, the co-located pixel group occupying a space in the target reference picture occupied at least in part by the current pixel group in the current picture;
• determining that a neighbor pixel group has an assigned reference vector that indicates a predictive relationship that points to a reference vector picture;
• scaling the predictive relationship according to a relative temporal position of the target reference picture, the reference vector picture, and the current picture to determine an offset vector;
• selecting a default motion vector for the current pixel group;
• applying the offset vector by shifting the current pixel group in accord with the offset vector, and then determining whether a pre-determined pixel of a pixel subgroup of the current pixel group occupies a spatial position that is occupied by a predictor pixel group of the target reference picture that is assigned a predictor motion vector;
• when the pre-determined pixel of the pixel subgroup occupies the spatial position that is occupied by the predictor pixel group of the target reference picture that is assigned the predictor motion vector, processing the predictor motion vector to generate a subgroup motion vector for the pixel subgroup;
• and
• when the pre-determined pixel of the pixel subgroup does not occupy the spatial position that is occupied by the predictor pixel group of the target reference picture that is assigned the predictor motion vector, using the default motion vector as the subgroup motion vector irrespective of whether other predictor pixel groups are assigned prediction motion vectors.

A product comprising:

• a machine readable medium other than a transitory signal;
• and
• instructions stored on the machine readable medium, the instructions, when executed, configured to cause motion vector prediction circuitry to:
• identify a candidate motion vector assigned to a co-located pixel group in a target reference picture for a current pixel group in a current picture, the co-located pixel group comprising a pixel position in the target reference picture included in the current pixel group in the current picture;
• in response to identifying the candidate motion vector, select the target reference picture;
• initiate processing of the candidate motion vector to generate a default motion vector;
• and
• determine subgroup motion vectors for multiple pixel subgroups of the current pixel group by:
• identifying a reference vector assigned to a neighbor pixel group, the reference vector configured to signal a predictive relationship that points to a reference vector picture;
• determining a scaling factor using a ratio of a first duration between the target reference picture and the current picture to a second duration between the target reference picture and the reference vector picture;
• applying the scaling factor to the reference vector to determine an offset vector;
• shifting the current pixel group in accord with the offset vector, and then determine whether a pre-determined pixel of the pixel subgroup occupies a spatial position that is occupied by a predictor pixel group that is assigned a predictor motion vector;
• when the pre-determined pixel of the pixel subgroup occupies the spatial position that is occupied by the predictor pixel group that is assigned the predictor motion vector, processing the predictor motion vector to generate a subgroup motion vector for the pixel subgroup;
• and
• when the pre-determined pixel of the pixel subgroup does not occupy the spatial position that is occupied by the predictor pixel group that is assigned the predictor motion vector, assigning the default motion vector to the pixel subgroup without searching other predictor pixel groups for a predictor motion vector for the pixel subgroup.