Diversified motion using multiple global motion models

A method for encoding a current frame of a video, comprising:

• assigning initial global motion models (GMMs) to reference frames;
• setting a current GMM set to the initial GMMs;
• partitioning the current frame into blocks;
• assigning, to each subset of the blocks, one of the initial GMMs;
• computing an error of encoding the current frame using the initial GMMs by predicting each block using the respective initial GMM assigned to the each block;
• until an objective function is met, iteratively performing:
• selecting one GMM of the current GMM set and fixing all other GMMs of the current GMM set;
• obtaining at least one candidate GMM for the one selected GMM;
• determining a respective error of encoding the current frame using candidate GMM sets, wherein each candidate GMM set includes the all other GMMs of the current GMM set and one of the at least one candidate GMM for the one selected GMM;
• and
• setting the current GMM set to the candidate GMM set corresponding to a lowest respective error;
• and
• encoding the current frame using the current GMM set.

A method for encoding a current frame of a video, comprising:

• jointly determining global motion models (GMMs) for reference frames, wherein jointly determining the GMMs for the reference frames comprises:
• until an objective function is met, performing steps including:
• refining respective GMMs of the reference frames assigned to respective groups of blocks of the current frame, wherein refining the respective GMMs of the reference frames comprises:
• assigning a first GMM of a first reference frame of the reference frames to a first group of blocks of the current frame;
• generating an updated first GMM using a subset of the first group of blocks, wherein the subset of the first group of blocks includes at least one less block than the first group of blocks;
• and
• assigning the updated first GMM to the first group of blocks;
• and
• updating an assignment of the respective GMMs to blocks of the current frame;
• and
• encoding the current frame using the GMMs.