Methods and apparatuses of video processing with motion refinement and sub-partition base padding

1. A method of processing video data in a video coding system, comprising:

receiving input data associated with a current block in a current picture;

partitioning the current block into sub-partitions;

for each sub-partition in the current block:

obtaining a reference block from one or more reference pictures according to an initial Motion Vector (MV);

deriving a refined MV by performing motion refinement on each sub-partition by searching around the initial MV with N-pixel refinement, wherein deriving the refined MV comprises applying an integer search followed by a fractional sample refinement, wherein the fractional sample refinement is derived using a parametric error surface operation, and a center position cost and costs of four neighboring positions are used to fit a two-dimensional parabolic error surface equation; and

padding one or more boundary pixels of the reference block;

performing motion compensations for the sub-partitions in the current block according to the refined MVs to generate a final predictor for the current block; and

encoding or decoding the current block according to the final predictor.

2. The method of claim 1 , further comprising checking if a size, width, or height of the current block is larger than or equal to a threshold, and partitioning the current block into sub-partitions when the size, width, or height of the current block is larger than or equal to the threshold.

3. The method of claim 2 , wherein the threshold is 16×16, 16×8, 8×16, 8 or 16, and a size of each sub-partition split from the current block is 16×16, 16×8, or 8×16 samples.

4. The method of claim 1 , wherein a size of each sub-partition is W×H samples and a size of the reference block for each sub-partition is (W+7)×(H+7) samples.

5. The method of claim 1 , wherein the motion refinement applied to each sub-partition includes an integer search with up to 2-pixel refinement in horizontal and vertical directions.

6. The method of claim 1 , wherein padding comprises extending the reference block for each sub-partition by repeating top, bottom, left, and right boundary pixels of the reference block.

7. The method of claim 1 , wherein an 8-connected search pattern is used in the integer search, and the 8-connected search pattern checks eight connected integer positions related to a central position including left, top, right, bottom, left-top, right-top, left-bottom, and right-bottom neighboring positions of the central position.

8. The method of claim 1 , wherein the parametric error surface operation is disabled when a later Bi-directional Optical Flow (BDOF) operation is enabled.

9. The method of claim 1 , wherein applying the integer search comprises employing a 2-pixel-distance MV change for one search iteration in the integer search.

10. The method of claim 1 , wherein performing motion refinement comprises determining a search region number N according to a size of the current block, or a frame resolution of the current picture.

11. The method of claim 1 , wherein obtaining the reference block for each sub-partition comprises interpolating the reference picture pixels into fractional position pixels if the initial MV has a fractional part.

12. The method of claim 1 , wherein obtaining the reference block for each sub-partition comprises directly using original integer pixels of the one or more reference pictures, and a parametric error surface operation is employed to decide a fractional pixel refinement.

13. The method of claim 12 , wherein the initial MV is rounded to a nearest integer MV, and the rounded MV is used to derive the reference block for each sub-partition.

14. The method of claim 1 , wherein padding one or more boundary pixels of the reference block for each sub-partition in the current block is only performed if any pixel needed for motion compensation of the sub-partition using the refined MV is outside the reference block.

15. The method of claim 1 , wherein performing motion refinement comprises performing only on one or more pre-defined or selected sub-partitions, a sub-partition other than the pre-defined or selected sub-partitions directly uses a neighboring sub-partition motion refinement result.

16. The method of claim 1 , wherein motion refinement is skipped for the current block if an initial Sum of Absolute Difference (SAD) of an initial matching position is larger than a pre-defined threshold.

17. The method of claim 1 , further comprising disabling a later Bi-directional Optical Flow (BDOF) operation if a MV difference between the initial MV and the refined MV is larger than a threshold.

18. An apparatus of processing video data in a video coding system, the apparatus comprising one or more electronic circuits configured for:

receiving input data associated with a current block in a current picture;

partitioning the current block into sub-partitions;

for each sub-partition in the current block:

obtaining a reference block from one or more reference pictures according to an initial Motion Vector (MV);

deriving a refined MV by performing motion refinement on each sub-partition by searching around the initial MV with N-pixel refinement, wherein deriving the refined MV comprises applying an integer search followed by a fractional sample refinement, wherein the fractional sample refinement is derived using a parametric error surface operation, and a center position cost and costs of four neighboring positions are used to fit a two-dimensional parabolic error surface equation; and

padding one or more boundary pixels of the reference block;

performing motion compensation for the sub-partitions in the current block according to the refined MVs to generate a final predictor for the current block; and

encoding or decoding the current block according to the final predictor.