Method and apparatus for video coding

1. A method for video encoding, comprising:

determining an inter prediction mode for a current block;

generating initial predictions for samples in a subblock of a plurality of subblocks in the current block based on the inter prediction mode; and

for one of the samples in the subblock,

determining a motion vector (MV) refinement (MVR), an absolute value of a first component of the MVR being one of: (i) 2 N1 and (ii) a sum or a difference of 2 N1 and 2 M1 , N1 and M1 being integers, the first component of the MVR being in a first direction that is one of an x direction and a y direction;

generating a prediction refinement based on the MVR and spatial gradients for the one of the samples, the spatial gradients being based on the initial predictions for the samples; and

encoding the current block in a bitstream based on the initial prediction of the one of the samples and the prediction refinement.

2. The method of claim 1 , wherein

the spatial gradients include a first spatial gradient in the first direction and a second spatial gradient in a second direction that is another one of the x direction and the y direction;

the MVR includes a second component that is in the second direction, an absolute value of the second component of the MVR being one of: (i) 2 N2 and (ii) a sum or a difference of 2 N2 and 2 M2 , N2 and M2 being integers; and

the generating the prediction refinement further includes:

determining a first refinement based on one of: (i) shifting the first spatial gradient by |N1| bits and (ii) separately shifting the first spatial gradient by |N1| and |M1| bits respectively and combining the shifted first spatial gradients;

determining a second refinement based on one of: (i) shifting the second spatial gradient by |N2| bits and (ii) separately shifting the second spatial gradient by |N2| and |M2| bits respectively and combining the shifted second spatial gradients; and

generating the prediction refinement based on the first refinement and the second refinement.

3. The method of claim 1 , wherein the determining the MVR further comprises:

determining an initial MVR for the one of the samples in the subblock based on the inter prediction mode; and

rounding a first component of the initial MVR to generate the first component of the MVR.

4. The method of claim 3 , wherein

the inter prediction mode is a subblock-based affine motion model with a prediction refinement with optical flow (PROF); and

for the one of the samples in the subblock, the determining the initial MVR further includes:

determining a sample MV of the one of the samples and a subblock MV of the subblock based on the subblock-based affine motion model; and

determining the initial MVR based the sample MV and the subblock MV.

5. The method of claim 3 , wherein the rounding the first component of the initial MVR further comprises:

rounding the first component of the initial MVR based on a position p of a most significant bit (MSB) of the first component of the initial MVR.

6. The method of claim 5 , wherein the rounding the first component of the initial MVR comprises:

left-shifting a bit value of 1 by p bits to obtain the absolute value of the first component of the MVR as 2 P , N1 being equal to p.

7. The method of claim 5 , wherein the rounding the first component of the initial MVR comprises:

left-shifting a bit value of 1 by p or (p+1) bits to round an absolute value of the first component of the initial MVR to a nearest power of 2 that is one of: 2 P and 2 P+1 .

8. The method of claim 3 , wherein

the first component of the initial MVR is a ratio of a first value over a second value, the first value and the second value being non-zero integers; and

the rounding the first component of the initial MVR further includes rounding the first component of the initial MVR based on a first position p1 of a MSB of the first value and a second position p2 of a MSB of the second value.

9. The method of claim 1 , wherein the first component of the MVR is 2 N1 or −2 N1 .

10. The method of claim 1 , wherein the first component of the MVR is (2 N1 +2 M1 ) or −(2 N1 −2 M1 ).

11. An apparatus for video encoding, comprising processing circuitry configured to:

determine an inter prediction mode for a current block;

generate initial predictions for samples in a subblock of a plurality of subblocks in the current block based on the inter prediction mode; and

for one of the samples in the subblock,

determine a motion vector (MV) refinement (MVR), an absolute value of a first component of the MVR being one of: (i) 2 N1 and (ii) a sum or a difference of 2 N1 and 2 M1 , N1 and M1 being integers, the first component of the MVR being in a first direction that is one of an x direction and a y direction;

generate a prediction refinement based on the MVR and spatial gradients for the one of the samples, the spatial gradients being based on the initial predictions for the samples; and

encode the current block in a bitstream based on the initial prediction of the one of the samples and the prediction refinement.

12. The apparatus of claim 11 , wherein

the spatial gradients include a first spatial gradient in the first direction and a second spatial gradient in a second direction that is another one of the x direction and the y direction;

the MVR includes a second component that is in the second direction, an absolute value of the second component of the MVR being one of: (i) 2 N2 and (ii) a sum or a difference of 2 N2 and 2 M2 , N2 and M2 being integers; and

the processing circuitry is further configured to:

determine a first refinement based on one of: (i) shifting the first spatial gradient by |N 1 | bits and (ii) separately shifting the first spatial gradient by |N 1 | and |M 1 | bits respectively and combining the shifted first spatial gradients;

determine a second refinement based on one of: (i) shifting the second spatial gradient by |N 2 | bits and (ii) separately shifting the second spatial gradient by |N 2 | and |M 2 | bits respectively and combining the shifted second spatial gradients; and

generate the prediction refinement based on the first refinement and the second refinement.

13. The apparatus of claim 11 , wherein the processing circuitry is further configured to:

determine an initial MVR for the one of the samples in the subblock based on the inter prediction mode; and

round a first component of the initial MVR to generate the first component of the MVR.

14. The apparatus of claim 13 , wherein

the inter prediction mode is a subblock-based affine motion model with a prediction refinement with optical flow (PROF); and

for the one of the samples in the subblock, the processing circuitry is configured to:

determine a sample MV of the one of the samples and a subblock MV of the subblock based on the subblock-based affine motion model; and

determine the initial MVR based the sample MV and the subblock MV.

15. The apparatus of claim 13 , wherein the processing circuitry is further configured to:

round the first component of the initial MVR based on a position p of a most significant bit (MSB) of the first component of the initial MVR.

16. The apparatus of claim 13 , wherein

the first component of the initial MVR is a ratio of a first value over a second value, the first value and the second value being non-zero integers; and

the processing circuitry is configured to round the first component of the initial MVR based on a first position p1 of a MSB of the first value and a second position p2 of a MSB of the second value.

17. The apparatus of claim 11 , wherein the first component of the MVR is 2 N1 or −2 N1 .

18. The apparatus of claim 11 , wherein the first component of the MVR is (2 N1 +2 M1 ) or −(2 N1 +2 M1 ).

19. A non-transitory computer-readable storage medium storing instructions which when executed by a processor cause the processor to perform:

determining an inter prediction mode for a current block;

generating initial predictions for samples in a subblock of a plurality of subblocks in the current block based on the inter prediction mode; and

for one of the samples in the subblock,

determining a motion vector (MV) refinement (MVR), an absolute value of a first component of the MVR being one of: (i) 2 N1 and (ii) a sum or a difference of 2 N1 and 2 M1 N1 and M1 being integers, the first component of the MVR being in a first direction that is one of an x direction and a y direction;

generating a prediction refinement based on the MVR and spatial gradients for the one of the samples, the spatial gradients being based on the initial predictions for the samples; and

encoding the current block in a bitstream based on the initial prediction of the one of the samples and the prediction refinement.

20. The non-transitory computer-readable storage medium of claim 19 , wherein the spatial gradients include a first spatial gradient in the first direction and a second spatial gradient in a second direction that is another one of the x direction and the y direction;

the MVR includes a second component that is in the second direction, an absolute value of the second component of the MVR being one of: (i) 2 N2 and (ii) a sum or a difference of 2 N2 and 2 M2 , N2 and M2 being integers; and

the generating the prediction refinement further includes:

determining a first refinement based on one of: (i) shifting the first spatial gradient by IN 11 bits and (ii) separately shifting the first spatial gradient by |N 1 | and |M 1 | bits respectively and combining the shifted first spatial gradients;

determining a second refinement based on one of: (i) shifting the second spatial gradient by |N 2 | bits and (ii) separately shifting the second spatial gradient by |N 2 | and |M 2 | bits respectively and combining the shifted second spatial gradients; and

generating the prediction refinement based on the first refinement and the second refinement.