Method and apparatus of bi-directional optical flow for video coding
Method and apparatus of using Bi-directional optical flow (BIO) for a true bi-direction predicted block are disclosed. According to one method of the present invention, the gradients are limited to a predefined bit-depth in order to reduce the computational complexity. According to another method, the data range of gradient sum and/or the difference between L0 and L1 interpolated pixels are shifted by a predefined bit-depth. The pre-defined bit-depth can be implicitly determined or signalled in a bit stream at a sequence level, picture level or slice level. The pre-defined bit-depth can also be determined dependent on input bit-depth.
1. A method of video coding using Bi-directional optical flow (BIO) for a true bi-direction predicted block, the method comprising:
receiving input data associated with a current block in a current picture, wherein the current block is coded using bi-direction prediction;
determining a first reference block associated with the current block in a first reference picture from reference picture list 0 and a second reference block associated with the current block in a second reference picture from reference picture list 1 , wherein the first reference block and the second reference block are pointed by true bi-direction motion vectors;
deriving first x-gradients and first y-gradients based on the first reference block, and second x-gradients and second y-gradients based on the second reference block,
wherein data range related to the first x-gradients, the first y-gradients, the second x-gradients and the second y-gradients are limited to a pre-defined bit-depth to generate range-limited first x-gradients, range-limited first y-gradients, range-limited second x-gradients and range-limited second y-gradients;
deriving a set of gradient parameters based on first information, wherein the first information comprising the range-limited first x-gradients, the range-limited first y-gradients, the range-limited second x-gradients, the range-limited second y-gradients;
deriving x-motion offset and y-motion offset based on the set of gradient parameters;
generating a refined predictor for the current block based on second information, wherein the second information comprising the x-motion offset and the y-motion offset; and
encoding or decoding the current block based on the refined predictor.
2. The method of claim 1 , wherein if the first x-gradients, the first y-gradients, the second x-gradients and the second y-gradients are greater than a maximum value or smaller than a minimum value, the first x-gradients, the first y-gradients, the second x-gradients and the second y-gradients are clipped to an allowable value indicated by the pre-defined bit-depth.
3. The method of claim 1 , wherein the pre-defined bit-depth is implicitly determined.
4. The method of claim 1 , wherein the pre-defined bit-depth is determined dependent on input bit-depth.
5. The method of claim 1 , wherein the pre-defined bit-depth is signalled in a bit stream at a sequence level, picture level or slice level.
6. A method of video coding using Bi-directional optical flow (BIO) for a true bi-direction predicted block, the method comprising:
receiving input data associated with a current block in a current picture, wherein the current block is coded using bi-direction prediction;
determining a first reference block associated with the current block in a first reference picture from reference picture list 0 and a second reference block associated with the current block in a second reference picture from reference picture list 1 , wherein the first reference block and the second reference block are pointed by true bi-direction motion vectors;
deriving first x-gradients and first y-gradients based on the first reference block, and second x-gradients and second y-gradients based on the second reference block;
deriving x-gradient sums based on the first x-gradients and the second x-gradients; deriving y-gradient sums based on the first y-gradients and the second y-gradients; and deriving reference differences between interpolated pixels in the first reference block and the second reference block; wherein a data range for the x-gradient sums and the y-gradient sums, the reference differences, or both are shifted by a first pre-defined bit-depth;
deriving x-motion offset and y-motion offset using first information, wherein the first information comprising the x-gradient sums, the y-gradient sums, and the reference differences;
generating a refined predictor for the current block based on second information, wherein the second information comprising the x-motion offset and the y-motion offset; and
encoding or decoding the current block based on the refined predictor.
7. The method of claim 6 , wherein the first pre-defined bit-depth is implicitly determined.
8. The method of claim 6 , wherein the first pre-defined bit-depth is determined dependent on input bit-depth.
9. The method of claim 6 , wherein the first pre-defined bit-depth is signalled in a bit stream at a sequence level, picture level or slice level.
10. The method of claim 6 , wherein the data range for the x-gradient sums and the y-gradient sums, the reference differences, or both are limited to a second pre-defined bit-depth prior to being shifted by the first pre-defined bit-depth.
11. The method of claim 6 , wherein the data range for the x-gradient sums and the y-gradient sums, the reference differences, or both are limited to a second pre-defined bit-depth after being shifted by the first pre-defined bit-depth.
12. An apparatus of video coding using Bi-directional optical flow (BIO) for a true bi-direction predicted block, the apparatus of video coding comprising one or more electronic circuits or processors arranged to:
receive input data associated with a current block in a current picture, wherein the current block is coded using bi-direction prediction;
determine a first reference block associated with the current block in a first reference picture from reference picture list 0 and a second reference block associated with the current block in a second reference picture from reference picture list 1 , wherein the first reference block and the second reference block are pointed by true bi-direction motion vectors;
derive first x-gradients and first y-gradients based on the first reference block, and second x-gradients and second y-gradients based on the second reference block;
derive x-gradient sums based on the first x-gradients and the second x-gradients; deriving y-gradient sums based on the first y-gradients and the second y-gradients; and deriving reference differences between interpolated pixels in the first reference block and the second reference block; wherein a data range for the x-gradient sums and the y-gradient sums, the reference differences, or both are shifted by a first pre-defined bit-depth;
derive x-motion offset and y-motion offset using first information, wherein the first information comprising the x-gradient sums, the y-gradient sums, and the reference differences;
generate a refined predictor for the current block based on second information, wherein the second information comprising the x-motion offset and the y-motion offset; and
encode or decode the current block based on the refined predictor.