Content adaptive prediction and entropy coding of motion vectors for next generation video

A video encoder comprising:

• an image buffer;
• a graphics processing unit comprising entropy encoder logic circuitry, wherein the graphics processing unit is communicatively coupled to the image buffer and wherein the entropy encoder logic circuitry is configured to:
• receive motion vector data comprising a plurality of motion vectors for a video frame;
• determine a first predicted motion vector using a median motion vector prediction technique for each motion compensated block of the video frame based on motion vectors of its neighboring blocks;
• determine a second predicted motion vector using a selected motion vector prediction technique for each motion compensated block based on motion vectors of its neighboring blocks, wherein the selected motion vector prediction technique is a first motion vector prediction technique when the median based motion vector is a zero motion vector and a second motion vector prediction technique different than the first motion vector prediction technique when the median based motion vector is a non-zero motion vector;
• select the first or the second predicted motion vector for each motion compensated block and define a motion vector bit mask identifying the median or the selected motion vector prediction technique for each motion compensated block;
• determine a selected coding method for the motion vector bit mask;
• and
• encode the motion vector bit mask into a bitstream based on the selected coding method.

A system comprising:

• a video decoder configured to decode an entropy encoded bitstream, wherein the video decoder is configured to:
• decode the entropy encoded bitstream to determine a frame level motion vector prediction method and a motion vector bit mask for a video frame, wherein the motion vector bit mask identifies a median or a selected motion vector prediction technique for each motion compensated block of the video frame based on motion vectors of its neighboring blocks, and wherein the selected motion vector prediction technique is based on the frame level motion vector prediction method and a first motion vector for each motion compensated block determined based on the median motion vector prediction technique;
• determine, based on the motion vector bit mask, a first individual motion compensated block is associated with the median motion vector prediction technique and a second individual motion compensated block is associated with the selected motion vector prediction technique;
• generate a first predicted motion vector associated with the first individual motion compensated block based on the median motion vector prediction technique;
• generate a median based prediction motion vector associated with the second individual motion compensated block;
• and
• generate a second predicted motion vector associated with the second motion compensated block based on a spiral motion vector prediction technique when the median based prediction motion vector is a zero motion vector and based on at least one of a Graham motion vector prediction technique or a Graham last block motion vector prediction technique when the median based prediction motion vector is a non-zero motion vector.