Huawei Technologies Co., Limited

Jeeva Raj A

Sriram Sethuraman

Sagar Kotecha

Semih Esenlik

Shan Liu

Haitao Yang

Huanbang Chen

Jiantong Zhou

Shan Gao

Xiang Ma

Jay Nitin Shingala

Given that decoder side motion vector refinement/derivation is a normative aspect of a coding system, the encoder will also have to perform the same error surface technique in order to not have any drift between the encoder's reconstruction and the decoder's reconstruction. Hence, all aspects of all embodiments are applicable to both encoding and decoding systems. In template matching, the refinement movement occurs only in the reference starting from the sub-pixel accurate center that is derived based on the explicitly signaled merge index or implicitly through cost evaluations. In bilateral matching (with or without averaged template), the refinements start in the reference lists L0 and L1 starting from the respective sub-pixel accurate centers that are derived based on the explicitly signaled merge index or implicitly through cost evaluations.

Error surface based sub-pixel accurate refinement method for decoder side motion vector refinement

A bidirectional optical flowing prediction method includes obtaining an initial motion vector pair for a current block, obtaining a forward and a backward prediction block according to the forward motion vector and a backward prediction block according to the initial motion vector pair, and calculating gradient parameters for a current sample in the current block. The method further includes obtaining at least two sample optical flow parameters, including a first parameter and a second parameter, for the current sample based on the gradient parameters, obtaining block optical flow parameters based on sample optical flow parameters of samples in the current block, and obtaining a prediction value of the current block. One of the block optical flow parameters is obtained by multiplying the first parameter and a sign function of the second parameter, and the sign function is a piecewise function with at least three subintervals.

Encoder, a decoder and corresponding methods for inter prediction using gradient parameters

A method for inter-prediction of a current image block in a current picture of a video is provided. The method includes determining whether a first temporal distance (such as TD0) is equal to a second temporal distance (such as TD1), wherein the first temporal distance is represented in terms of a difference between a picture order count value of the current picture and a picture order count value of a first reference picture; and the second temporal distance is represented in terms of a difference between a picture order count value of a second reference picture and the picture order count value of the current picture; and performing no motion vector refinement (DMVR) procedure when it is determined that the first temporal distance (TD0) is not equal to the second temporal distance (TD1). Thus the DMVR procedure is restricted to only the image block with equal-distance references.

Apparatus and method for conditional decoder-side motion vector refinement in video coding

A method for determining a motion vector to be used in inter-prediction of a current block of a video picture, the method comprising: obtaining an initial motion vector; obtaining at least two groups of points according to the initial motion vector, wherein a first group of the at least two groups of points only includes a first point corresponding to the initial motion vector and additional N points with N≥0 and wherein each point of the first group is closer to the first point than any point of the points of a second group of the at least two groups; and obtaining the motion vector for the current block according to the at least two groups of points and a cost function.

Search region for motion vector refinement

In some embodiments, a video encoder or a video decoder obtains a pair of motion vectors for a current coding block of a video signal with respect to two reference frames. Predictions of the current block are generated using the pair of motion vectors from respective reference frames. The video encoder or video decoder further determine an optical flow for the current coding block based on samples values in the predictions. One component of the optical flow is determined based on the other component of the optical flow. Bi-predictive optical flow (BPOF) can be performed on the current coding block using the determined optical flow.

Optical flow based video inter prediction

It is provided a method of video coding implemented in a decoding device or an encoding device, the method comprising: obtaining initial motion vectors for a current block; obtaining first predictions for a sample value in the current block based on the initial motion vectors; calculating a first matching cost according to the first predictions; determining whether an optical flow refinement process should be performed or not, according to at least one preset condition, the at least one preset condition comprising a condition of whether the calculated first matching cost is equal to or larger than a threshold value; and performing an optical flow refinement process for obtaining a final inter prediction for the sample value in the current block, when it is determined that the optical flow refinement process should be performed.

Early termination for optical flow refinement

A method for bi-directional optical flow prediction based on reference sample values from sample positions belonging to an extended region of a predicted block used in the bi-directional optical flow prediction of a current block is provided, wherein sub-pixel positions in the extended region of the predicted block in the reference picture are rounded to obtain integer pixel sample positions in the reference picture, wherein the reference sample values at the obtained integer pixel sample positions in the reference picture are used in the computation of boundary sample gradients corresponding to the boundaries of the predicted block in the reference picture.

Device and method for computing position of integer grid reference sample for block level boundary sample gradient computation

The present disclosure provides an inter prediction method, comprising the steps of obtaining an initial motion vector and a reference picture for bi-prediction; obtaining sets of candidate sample positions in the reference picture according to the initial motion vector and candidate motion vectors, wherein each candidate motion vector is derived by the initial motion vector and a respective motion vector offset, and wherein each set of candidate sample positions corresponds to each candidate motion vector; obtaining a respective set of sample positions from each set of candidate sample positions; computing a matching cost for each candidate motion vector within each set of sample positions; obtaining a refined motion vector based on the computed matching cost of each candidate motion vector; and obtaining prediction values for a current block based on the refined motion vector.

DMVR using decimated prediction block

Encoder, a decoder and corresponding methods for inter prediction using bidirectional optical flow

A method for inter prediction of a block of a picture and corresponding apparatus are provided. The method comprises: determining a first prediction block based on an initial first motion vector of a current block in a first reference picture, so as to obtain a first extension block; obtaining from the first reference picture, one or more first reference blocks each of which is associated with a combination of the initial first motion vector and one of motion vector offsets; obtaining a target motion vector offset with a smallest cost among the motion vector offsets; determining for the current block, a refined first motion vector which is a combination of the initial first motion vector and the target motion vector offset; and obtaining a refined prediction block of the current block based on the refined first motion vector, thereby improving coherence of refined motion vectors to achieve higher coding efficiencies.

Use of extended samples during search in decoder-side motion refinement

Methods and apparatuses of determining an alignment level between motion compensated reference patches for reducing motion vector refinement steps are provided. According to one method, obtaining, by a decoder, motion compensated interpolated samples based on sub-pixel accurate merge motion vectors from a bilinear motion compensated interpolation; computing, by the decoder, a sum of absolute differences (SAD) between two motion compensated reference patches using a subset of the motion compensated interpolated samples; determining, by the decoder, whether the SAD is less than a coding unit (CU) size-dependent threshold value; when the SAD is less than the CU size-dependent threshold value: skipping remaining decoder-side motion vector refinement (DMVR) process steps; and performing final motion compensation; and when the SAD is not less than the CU size-dependent threshold value: performing the remaining DMVR process steps; and performing the final motion compensation.

Growing community of inventors

Karnataka, India

Jeeva Raj A