Sub-PU based bi-directional motion compensation in video coding

A method of encoding video data, the method comprising:

• partitioning, into a plurality of sub-blocks, a block of a picture of the video data;
• and
• for each respective sub-block of the plurality of sub-blocks:
• deriving a respective first motion vector of the respective sub-block based on motion information for at least two blocks neighboring the respective sub-block, wherein, for a particular sub-block of the plurality of sub-blocks, deriving the respective first motion vector of the particular sub-block comprises determining the respective first motion vector of the particular sub-block as a weighted sum of a motion vector of an above neighbor block for the particular sub-block, a motion vector of a left neighbor block for the particular sub-block, and a scaled motion vector of a motion vector of a block of a reference picture collocated with a below-right neighbor block for the particular sub-block;
• determining, based on a coding block for the respective sub-block, a second motion vector for the respective sub-block that corresponds to a motion vector for a reference picture for the respective sub-block;
• determining, based on a difference between the coding block for the respective sub-block and the second motion vector of the respective sub-block, a respective motion vector difference for the respective sub-block;
• signaling the respective motion vector difference for the respective sub-block in a bitstream;
• and
• generating, based on the first motion vector of the respective sub-block and the second motion vector of the respective sub-block, a predictive block for the respective sub-block.

An apparatus for decoding video data comprising:

• one or more storage media configured to store video data;
• and
• one or more processors configured to:
• partition, into a plurality of sub-blocks, a block of a picture of the video data;
• and
• for each respective sub-block of the plurality of sub-blocks:
• derive a respective first motion vector of the respective sub-block based on motion information for at least two blocks neighboring the respective sub-block, wherein, for a particular sub-block of the plurality of sub-blocks, to derive the respective first motion vector of the particular sub-block, the one or more processors are configured to determine the respective first motion vector of the particular sub-block as a weighted sum of a motion vector of an above neighbor block for the particular sub-block, a motion vector of a left neighbor block for the particular sub-block, and a scaled motion vector of a motion vector of a block of a reference picture collocated with a below-right neighbor block for the particular sub-block;
• determine, based on a respective motion vector difference for the respective sub-block signaled in a bitstream, a second motion vector for the respective sub-block;
• and
• generate, based on the first motion vector of the respective sub-block and the second motion vector of the respective sub-block, a predictive block for the respective sub-block.

An apparatus for encoding video data comprising:

• one or more storage media configured to store video data;
• and
• one or more processors configured to:
• partition, into a plurality of sub-blocks, a block of a picture of the video data;
• and
• for each respective sub-block of the plurality of sub-blocks:
• derive a respective first motion vector of the respective sub-block based on motion information for at least two blocks neighboring the respective sub-block, wherein, for a particular sub-block of the plurality of sub-blocks, to derive the respective first motion vector of the particular sub-block, the one or more processors are configured to determine the respective first motion vector of the particular sub-block as a weighted sum of a motion vector of an above neighbor block for the particular sub-block, a motion vector of a left neighbor block for the particular sub-block, and a scaled motion vector of a motion vector of a block of a reference picture collocated with a below-right neighbor block for the particular sub-block;
• determine, based on a coding block for the respective sub-block, a second motion vector for the respective sub-block that corresponds to a motion vector for a reference picture for the respective sub-block;
• determine, based on a difference between the coding block for the respective sub-block and the second motion vector of the respective sub-block, a respective motion vector difference for the respective sub-block;
• signal the respective motion vector difference for the respective sub-block in a bitstream;
• and
• generate, based on the first motion vector of the respective sub-block and the second motion vector of the respective sub-block, a predictive block for the respective sub-block.