Affine motion model derivation method
Systems and methods are described for video coding using affine motion prediction. In an example method, motion vector gradients are determined from respective motion vectors of a plurality of neighboring sub-blocks neighboring a current block. An estimate of at least one affine parameter for the current block is determined based on the motion vector gradients. An affine motion model is determined based at least in part on the estimated affine parameter(s), and a prediction of the current block is generated using the affine motion model. The estimated parameter(s) may be used in the affine motion model itself. Alternatively, the estimated parameter(s) may be used in a prediction of the affine motion model. In some embodiments, only neighboring sub-blocks above and/or to the left of the current block are used in estimating the affine parameter(s).
1. A video decoding method comprising:
for at least one current block in a video, determining at least one motion vector gradient from respective motion vectors of a plurality of neighboring sub-blocks neighboring the current block;
based at least in part on the motion vector gradient, selecting a motion model for prediction of the current block, wherein the selection is made from among a set of motion models including at least a four-parameter affine motion model and a six-parameter affine motion model; and
generating a prediction of the current block using the selected motion model.
2. The method of claim 1 , wherein the set of motion models further includes a translational motion model.
3. The method of claim 1 , wherein the selection of the motion model for prediction of the current block is based on a model fitting error.
4. The method of claim 1 , wherein selecting a motion model comprises, for each of a plurality of candidate motion models from among the set of motion models:
deriving estimated motion vectors for a plurality of the neighboring sub-blocks from the respective motion model, and determining a model fitting error based on differences between actual motion vectors of the plurality of neighboring sub-blocks and the corresponding estimated motion vectors of the plurality of neighboring sub-blocks;
wherein the selected motion model is selected to minimize the model fitting error.
5. The method of claim 4 , wherein the model fitting error is a weighted sum of differences between the actual motion vectors of the neighboring sub-blocks and the corresponding estimated motion vectors of the respective neighboring sub-blocks, each difference in the sum being weighed based on a distance between the current block and the respective neighboring sub-block.
6. A video decoder apparatus comprising one or more processors configured to perform at least:
for at least one current block in a video, determining at least one motion vector gradient from respective motion vectors of a plurality of neighboring sub-blocks neighboring the current block;
based at least in part on the motion vector gradient, selecting a motion model for prediction of the current block, wherein the selection is made from among a set of motion models including at least a four-parameter affine motion model and a six-parameter affine motion model; and
generating a prediction of the current block using the selected motion model.
7. The apparatus of claim 6 , wherein the set of motion models further includes a translational motion model.
8. The apparatus of claim 6 , wherein the selection of the motion model for prediction of the current block is based on a model fitting error.
9. The apparatus of claim 6 , wherein selecting a motion model comprises, for each of a plurality of candidate motion models from among the set of motion models:
deriving estimated motion vectors for a plurality of the neighboring sub-blocks from the respective motion model, and determining a model fitting error based on differences between actual motion vectors of the plurality of neighboring sub-blocks and the corresponding estimated motion vectors of the plurality of neighboring sub-blocks;
wherein the selected motion model is selected to minimize the model fitting error.
10. The apparatus of claim 9 , wherein the model fitting error is a weighted sum of differences between the actual motion vectors of the neighboring sub-blocks and the corresponding estimated motion vectors of the respective neighboring sub-blocks, each difference in the sum being weighed based on a distance between the current block and the respective neighboring sub-block.
11. A video encoding method comprising:
for at least one current block in a video, determining at least one motion vector gradient from respective motion vectors of a plurality of neighboring sub-blocks neighboring the current block;
based at least in part on the motion vector gradient, selecting a motion model for prediction of the current block, wherein the selection is made from among a set of motion models including at least a four-parameter affine motion model and a six-parameter affine motion model; and
generating a prediction of the current block using the selected motion model.
12. The method of claim 11 , wherein the set of motion models further includes a translational motion model.
13. The method of claim 11 , wherein the selection of the motion model for prediction of the current block is based on a model fitting error.
14. The method of claim 11 , wherein selecting a motion model comprises, for each of a plurality of candidate motion models from among the set of motion models:
deriving estimated motion vectors for a plurality of the neighboring sub-blocks from the respective motion model, and determining a model fitting error based on differences between actual motion vectors of the plurality of neighboring sub-blocks and the corresponding estimated motion vectors of the plurality of neighboring sub-blocks;
wherein the selected motion model is selected to minimize the model fitting error.
15. The method of claim 14 , wherein the model fitting error is a weighted sum of differences between the actual motion vectors of the neighboring sub-blocks and the corresponding estimated motion vectors of the respective neighboring sub-blocks, each difference in the sum being weighed based on a distance between the current block and the respective neighboring sub-block.
16. A video encoder apparatus comprising one or more processors configured to perform at least:
for at least one current block in a video, determining at least one motion vector gradient from respective motion vectors of a plurality of neighboring sub-blocks neighboring the current block;
based at least in part on the motion vector gradient, selecting a motion model for prediction of the current block, wherein the selection is made from among a set of motion models including at least a four-parameter affine motion model and a six-parameter affine motion model; and
generating a prediction of the current block using the selected motion model.
17. The apparatus of claim 16 , wherein the set of motion models further includes a translational motion model.
18. The apparatus of claim 16 , wherein the selection of the motion model for prediction of the current block is based on a model fitting error.
19. The apparatus of claim 16 , wherein selecting a motion model comprises, for each of a plurality of candidate motion models from among the set of motion models:
deriving estimated motion vectors for a plurality of the neighboring sub-blocks from the respective motion model, and determining a model fitting error based on differences between actual motion vectors of the plurality of neighboring sub-blocks and the corresponding estimated motion vectors of the plurality of neighboring sub-blocks;
wherein the selected motion model is selected to minimize the model fitting error.
20. The apparatus of claim 19 , wherein the model fitting error is a weighted sum of differences between the actual motion vectors of the neighboring sub-blocks and the corresponding estimated motion vectors of the respective neighboring sub-blocks, each difference in the sum being weighed based on a distance between the current block and the respective neighboring sub-block.