Motion vector prediction design simplification

The present application claims priority to:

U.S. Provisional App. No. 61/504,085 for “Method for Simplification of MVP Design for HEVC” filed Jul. 1, 2011;

U.S. Provisional App. No. 61/508,844 for “Simplification of MVP Design for HEVC” filed Jul. 18, 2011;

U.S. Provisional App. No. 61/557,411 for “Simplification of MVP Design for HEVC” filed Nov. 8, 2011; and

U.S. Provisional App. No. 61/557,418 for “Parallel Friendly MVP Candidate Calculation for HEVC” filed Nov. 9, 2011, the contents of all of which are incorporated herein by reference in their entirety.

Motion Vector Prediction Design

Video compression systems employ block processing for most of the compression operations. A block is a group of neighboring pixels and may be treated as one coding unit in terms of the compression operations. Theoretically, a larger coding unit is preferred to take advantage of correlation among immediate neighboring pixels. Various video compression standards, e.g., Motion Picture Expert Group (MPEG)-1, MPEG-2, and MPEG-4, use block sizes of 4×4, 8×8, and 16×16 (referred to as a macroblock (MB)).

High efficiency video coding (HEVC) is also a block-based hybrid spatial and temporal predictive coding scheme. HEVC partitions an input picture into square blocks referred to as largest coding units (LCUs) as shown in FIG. 1 . Unlike prior coding standards, the LCU can be as large as 128×128 pixels. Each LCU can be partitioned into smaller square blocks called coding units (CUs). FIG. 2 shows an example of an LCU partition of CUs. An LCU 100 is first partitioned into four CUs 102 . Each CU 102 may also be further split into four smaller CUs 102 that are a quarter of the size of the CU 102 . This partitioning process can be repeated based on certain criteria, such as limits to the number of times a CU can be partitioned may be imposed. As shown, CUs 102 - 1 , 102 - 3 , and 102 - 4 are a quarter of the size of LCU 100 . Further, a CU 102 - 2 has been split into four CUs 102 - 5 , 102 - 6 , 102 - 7 , and 102 - 8 .

Each CU 102 may include one or more prediction units (PUs). FIG. 3 shows an example of a CU partition of PUs. The PUs may be used to perform spatial prediction or temporal prediction. A CU can be either spatially or temporally predictive coded. If a CU is coded in intra mode, each PU of the CU can have its own spatial prediction direction. If a CU is coded in inter mode, each PU of the CU can have its own motion vector(s) and associated reference picture(s).

In HEVC, motion vectors (MVs) are predictively coded. For a current PU having one current motion vector and an associated reference index, a motion vector predictor (MVP) is derived from motion vectors of spatially neighboring or temporally co-located PUs of the current PU. The difference between the current motion vector and the MVP is then determined and coded. This reduces overhead as only the difference is sent instead of information for the current motion vector.

Advanced motion vector prediction (AMVP) proposes to select the best MVP for a current motion vector from a candidate set of MVPs. The candidate set of MVPs for the current motion vector of the current PU is composed of motion vectors of spatial neighboring and co-located PUs. The candidate set of MVPs may be reduced by removing duplicated MVP candidates. An index indicating the selected MVP candidate, mvpIdx, may be transmitted in the bitstream to indicate the selected MVP.

In bi-prediction mode, the current PU may have two motion vectors, a motion vector MV 0 for a reference list 0 and a motion vector MV 1 for a reference list 1 . The corresponding reference pictures pointed to by the two motion vectors MV 0 and MV 1 are indexed by reference list 0 and reference list 1 , respectively. To code the two motion vectors, two MVP candidate sets are first formed. One candidate set is for the list 0 motion vector and the other candidate set is for the list 1 motion vector. The same procedure described above applies to find the best MVPs for motion vectors MV 0 and MV 1 . Thus, two searches of two candidate lists are performed. The determination of two MVPs for two lists increases the computational process. Two MVP indices may also be then transmitted in the bitstream, which increases overhead.

Parallel Processing of MVPs for PUs

A motion vector predictor (MVP) of a current PU is derived from motion vectors (MV) of the current PU's spatial and temporal (collocated) neighboring blocks.

FIGS. 13A and 13B show two examples where in (a), a CU contains only one PU of the same size, and in (b), a CU has two non-square PUs (PU 0 and PU 1 ). In some examples, the spatial neighboring blocks for a current PU may or may not be in the same CU where the current PU block resides. In FIG. 13A , all the spatial neighboring blocks (shown with diagonal lines) of a current PU are outside of the current CU. On the other hand, in FIG. 13B , the spatial neighboring blocks for PU 0 are all outside in the current CU. But, the left spatial neighboring block for PU 1 is PU 0 , which is inside the same current CU.

FIGS. 14A and 14B show two examples where a CU includes two PUs. In FIG. 14A , the right PU (PU 1 ) will use the left PU (PU 0 ) as one of the left spatial neighbors and in FIG. 14B , the bottom PU (PU 1 ) will use the top PU (PU 0 ) as one of the above spatial neighbors. As a result, the two PUs within a CU cannot be parallel processed; i.e., PU 1 has to wait for PU 0 to finish motion vector derivation before PU 1 can start its own MVP derivation.

Also, a fixed order for encoding possible candidate set is used. The index number for the left spatial neighbor is always smaller than that for the above spatial neighbor no matter what kind of PU partition is. This may result in a poor coding performance for some specific PU partitions.

In one embodiment, a method receives a unit of video content. The unit of video content is coded in a bi-prediction mode. A motion vector predictor candidate set is determined for a first motion vector for the unit. The method then determines a first motion vector predictor from the motion vector prediction candidate set for the first motion vector and calculates a second motion vector predictor for a second motion vector for the unit of video content. The second motion vector predictor is calculated based on the first motion vector or the first motion vector predictor.

In one embodiment, a method is provided comprising: receiving a first unit of video content; receiving a second unit of video content, wherein the first unit of video content and the second unit of video content are within a third unit of video content; determining, by a computing device, a first motion vector predictor for a first motion vector of the first unit of video content; calculating, by the computing device, a first motion vector predictor for the second unit of video content from the first motion vector predictor or the first motion vector of the first unit of video content.

In one embodiment, an apparatus is provided comprising: one or more computer processors; and a computer-readable storage medium comprising instructions, that when executed, control the one or more computer processors to be operable for: receiving a unit of video content, wherein the unit of video content is coded in a bi-prediction mode; determining a motion vector predictor candidate set for a first motion vector for the unit; determining a first motion vector predictor from the motion vector prediction candidate set for the first motion vector; and calculating a second motion vector predictor for a second motion vector for the unit of video content, the second motion vector predictor calculated based on the first motion vector or the first motion vector predictor.

In one embodiment, an apparatus is provided comprising: one or more computer processors; and a computer-readable storage medium comprising instructions, that when executed, control the one or more computer processors to be operable for: receiving a first unit of video content; receiving a second unit of video content, wherein the first unit of video content and the second unit of video content are within a third unit of video content; determining, by a computing device, a first motion vector predictor for a first motion vector of the first unit of video content; calculating, by the computing device, a first motion vector predictor for the second unit of video content from the first motion vector predictor or the first motion vector of the first unit of video content.

The following detailed description and accompanying drawings provide a better understanding of the nature and advantages of particular embodiments.

FIG. 1 shows an input picture partitioned into square blocks referred to as largest coding units (LCUs).

FIG. 2 shows an example of an LCU partition of CUs.

FIG. 3 shows an example of a CU partition of PUs.

FIG. 4 depicts an example of a system for encoding and decoding video content according to one embodiment.

FIG. 5 shows an example where both motion vectors MV 0 and MV 1 point to past reference pictures according to one embodiment.

FIG. 6 shows another example where motion vector MV 0 points to a past reference picture and motion vector MV 1 points to a future reference picture according to one embodiment.

FIG. 7 shows an example where motion vectors MV 0 and MV 1 point to past reference pictures according to one embodiment.

FIG. 8 shows a second example where motion vector MV 0 points to a past reference picture and motion vector MV 1 points to a future reference picture according to one embodiment.

FIG. 9 depicts a simplified flowchart of a method for determining a second motion vector predictor according to one embodiment.

FIG. 10A and FIG. 10B depict an example of calculating first and second motion vector predictor MVP_list 0 or/and MVP_list 1 of the second PU based upon the information of the first PU where multiple PUs exist within a CU according to one embodiment.

FIG. 11 depicts a simplified flowchart of a method for determining MVPs when two PUs are found in a coding unit according to one embodiment.

FIG. 12A depicts an example of encoder according to one embodiment.

FIG. 12B depicts an example of decoder according to one embodiment.

FIGS. 13A and 13B show two examples where in (a), a CU contains only one PU of the same size, and in (b), a CU has two non-square PUs (PU 0 and PU 1 ).

FIGS. 14A and 14B show two examples where a CU includes two PUs.

FIGS. 15A and 15B depict an example of blocks to use as candidates according to one embodiment.

Described herein are techniques for a video compression system. In the following description, for purposes of explanation, numerous examples and specific details are set forth in order to provide a thorough understanding of particular embodiments. Particular embodiments as defined by the claims may include some or all of the features in these examples alone or in combination with other features described below, and may further include modifications and equivalents of the features and concepts described herein.

Motion Vector Prediction Design

FIG. 4 depicts an example of a system 400 for encoding and decoding video content according to one embodiment. An encoder 402 includes a motion vector predictor (MVP) manager 406 - 1 and a decoder 404 includes an MVP manager 406 - 2 . MVP managers 406 perform motion vector prediction to determine an MVP for a unit of video content, such as a PU or other block size. In one embodiment, MVP managers 406 determines two motion vectors predictor for the PU. Motion vectors MV 0 and MV 1 for a current picture (e.g., the PU) point to two reference pictures (or blocks) indexed by a first reference list 0 and a second reference list 1 .

Particular embodiments derive a second motion vector predictor from another motion vector or another motion vector predictor. For example, MVP manager 406 first obtains a first motion vector predictor MVP_list 0 for motion vector MV 0 . This may be calculated as described above by determining motion vectors for neighboring PUs and co-located PUs. It should be noted that the discussion will describe first motion vector predictor MVP_list 0 as being first obtained, but second motion vector predictor MVP_list 1 may be first obtained. After obtaining first motion vector predictor MVP_list 0 , MVP manager 406 calculates a second motion vector predictor MVP_list 1 from either motion vector MV 0 or first motion vector predictor MVP_list 0 . For example, as will be described below, scaling is used to calculate second motion vector predictor MVP_list 1 . The scaling may be based on different characteristics, such as temporal distance or picture order count (POC).

In one embodiment, MVP manager 406 determines the MVP candidate set for motion vector MV 0 . The MVP candidate set may be determined using motion vectors for neighboring PUs and co-located PUs. MVP manager 406 then selects the best MVP candidate, first motion vector predictor MVP_list 0 and corresponding index mvpIdx_list 0 , for motion vector MV 0 from the MVP candidate set. The reference for current motion vector MV 0 is used to identify which neighbor (spatial or temporal neighbor) PU is selected as first motion vector predictor MVP_list 0 . In one embodiment, first motion vector predictor MVP_list 0 is also scaled if reference list 0 of motion vector MV 0 is different with mvpIdx_list 0 so that first motion vector predictor MVP_list 0 points to the same reference picture as motion vector MV 0 does. First motion vector predictor MVP_list 0 below will be supposed that it is already scaled. Once the neighbor PU is selected, the reference for the neighbor PU is also known. Therefore, the mvpIdx_list 0 is derived from the information of current PU.

MVP manager 406 then calculates the second MVP candidate, second motion vector predictor MVP_list 1 , for motion vector MV 1 from either motion vector MV 0 or the first motion vector predictor MVP_list 0 . The calculation is performed instead of selecting from a candidate set.

Temporal Distance

In one embodiment, MVP manager 406 - 1 uses temporal distance to calculate the second motion vector predictor MVP_list 1 for motion vector MV 1 . A first temporal distance TR 1 is between the current picture and the reference picture pointed to by motion vector MV 0 . A second temporal distance TR 2 is between the current picture and the reference picture pointed to by motion vector MV 1 . The second motion vector predictor MVP_list 1 for motion vector MV 1 is calculated from motion vector MV 0 . For example, the following may be used to calculate the second motion vector predictor MVP_list 1 as follows:

MVP_list1=(sign)(TR2/TR1)*MV0

where if motion vector MV 0 and motion vector MV 1 both point to the past or the future, the sign is positive (+) or otherwise, the sign is negative (−). The reason the sign is used is because motion vectors MV 0 and MV 1 may be pointing in opposite directions and thus a negative sign needs to be applied to the scaling.

FIG. 5 shows an example where both motion vectors MV 0 and MV 1 point to past reference pictures according to one embodiment. A current picture 502 includes the PU being encoded. Current picture 502 is shown along with first reference picture 504 - 1 and second reference picture 504 - 2 . Reference pictures 504 - 1 and 504 - 2 are in the past (have a lower display order) as compared to current picture 502 . A temporal interval is shown as TR 1 =0 and TR 2 =8. This may indicate that reference picture 504 - 1 is four pictures in the past and reference picture 504 - 2 is eight pictures in the past. Other temporal distances may also be measured.

The second motion vector predictor MVP_list 1 for motion vector MV 1 may be calculated as:

MVP_list1=(TR2/TR1)*MV0=(8/4)*MV0=MV0<<1

In this case, the second motion vector predictor MVP_list 1 is calculated based on the temporal distance ratio of 8/4*motion vector MV 0 . The scaling of motion vector MV 0 is used such that the second motion vector predictor MVP_list 1 is shown as pointing to reference picture 504 - 2 . The difference between motion vector MV 1 and the second motion vector predictor MVP_list 1 can then be calculated and coded.

FIG. 6 shows another example where motion vector MV 0 points to a past reference picture and motion vector MV 1 points to a future reference picture according to one embodiment. The temporal difference TR 1 is an interval of 4 pictures and the temporal distance TR 2 is an interval of 8 pictures.

The second motion vector predictor MVP_list 1 for motion vector MV 1 can be determined as follows:

MVP_list1=−(TR2/TR1)*MV0=−(8/4)*MV0=−(MV0<<1)

In this case, the second motion vector predictor MVP_list 1 is equal to the negative of the scaling of motion vector MV 0 . This is because motion vector MV 1 points to a future reference picture and not a past reference picture. As shown, the scaling of motion vector MV 0 is used such that the second motion vector predictor MVP_list 1 points to reference picture 504 - 2 . In using first motion vector MV 0 to calculate the second motion vector predictor MVP_list 1 , computational complexity is reduced.

Because a first motion vector predictor MVP_list 0 is available at both encoder 402 and decoder 404 , the first motion vector predictor MVP_list 0 may be used instead of motion vector MV 0 to calculate the second motion vector predictor MVP_list 1 . The following may be used to calculate the second motion vector predictor MVP_list 1 as follows:

MVP_list1=(sign)(TR2/TR1)*MVP_list0

where TR 1 is the temporal distance between the current picture and the reference picture pointed to by motion vector MV 0 (in one example, first motion vector predictor MVP_list 0 has been normalized by using the reference index of motion vector MV 0 . In other words, motion vector MV 0 and first motion vector predictor MVP_list 0 point to the same reference picture) and TR 2 is the temporal distance between the current picture and the reference picture pointed to by motion vector MV 1 . The sign is positive if the first motion vector predictor MVP_list 0 and motion vector MV 1 point to the past or the future. Otherwise, the sign is negative. In using first motion vector predictor MVP_list 0 to calculate the second motion vector predictor MVP_list 1 , computational complexity is reduced.

Accordingly, the second motion vector predictor MVP_list 1 is calculated from either motion vector MV 0 or the first motion vector predictor MVP_list 0 instead of selecting the MVP candidate from an MVP candidate set. Thus, there is no overhead required for coding and sending an MVP index for the second list, which results in a savings in bits. Also, there is no need to implement the MVP selection procedure to determine the second MVP candidate, MVP_list 1 . As a result, the complexity of particular embodiments is reduced in half as compared to using an MVP selection procedure for list 1 .

When encoder 402 encodes the bitstream for a current PU, MVP manager 406 - 1 determines the first motion vector predictor MVP_list 0 using the index sent in the bitstream for first motion vector predictor MVP_list 0 . Then, MVP manager 406 - 1 may calculate the second motion vector predictor MVP_list 1 . For example, MVP manager 406 - 1 may use the methods described above to calculate the second motion vector predictor MVP_list 1 . Encoder 402 then determines the differences between the first motion vector MV 0 and the first motion vector predictor MVP_list 0 and the second motion vector MV 1 and the second motion vector predictor MVP_list. Encoder 402 then sends the differences in the bitstream to decoder 404 .

When decoder 404 decodes the bitstream for a current PU, MVP manager 406 - 2 determines the first motion vector predictor MVP_list 0 using the index sent in the bitstream for first motion vector predictor MVP_list 0 . Then, MVP manager 406 - 2 may calculate the second motion vector predictor MVP_list 1 . For example, MVP manager 406 - 2 may use the methods described above to calculate the second motion vector predictor MVP_list 1 . Decoder 404 then determines the first motion vector MV 0 using the differences sent in the bitstream using first motion vector predictor MVP_list 0 and the second motion vector MV 1 using the differences sent in the bitstream using the calculated motion vector predictor MVP_list 1 . Second motion vector predictor MVP_list 1 may be a scaled version of what is calculated from first motion vector MV 0 or first motion vector predictor MVP_list 0 . The motion vectors MV 0 and MV 1 may then be used in a temporal prediction process.

Other examples for calculating the second motion vector predictor MVP_list 1 may be used. Other equations may also be used. For example, the following may be used:

MVP_list1=MVP0

MVP_list1=MVP_list0

Other equations may also be used. Encoder 402 and decoder 404 may select the calculation method that is used. For example, one PU may use scaling and another may use equation MVP_list 1 =MVP 0 .

MVP Candidate Calculation Using POC

In one embodiment, the picture order count (POC) may be used to derive the second motion vector predictor MVP_list 1 for motion vector MV 1 . For a given PU, the POC for a current picture and reference pictures may be poc_curr, poc_ref_list 0 , and poc_ref_list 1 , respectively. The POC difference diff_poc_curr_list 0 between the current picture and the reference picture indexed by first reference list ref_list 0 is as follows:

diff_poc_curr_list0=poc_curr−poc_ref_list0

where diff_poc_curr_list 0 is the difference between the current POC and the POC of the first reference picture in list 0 .

Similarly, the POC difference diff_poc_curr_list 1 between a current picture and a reference picture indexed by the second reference list ref_list 1 is defined as follows:

diff_poc_curr_list1=poc_curr−poc_ref_list1

where diff_poc_curr_list 1 is the difference between the POC of the current picture and the POC of the second reference picture list 1 .

Particular embodiments determine the second motion vector predictor MVP_list 1 for motion vector MV 1 by scaling motion vector MV 0 . An example calculation can be as follows:

MVP_list1=(diff_poc_curr_list1/diff_poc_curr_list0)*MV0

FIG. 7 shows an example where motion vectors MV 0 and MV 1 point to past reference pictures according to one embodiment. In FIG. 7 , a current picture 502 has a POC of 8. Motion vector MV 0 points to a reference picture 504 - 1 that has a POC of 4 and a motion vector MV 1 points to a reference picture 504 - 2 that has a POC of 0. The second motion vector predictor MVP_list 1 for motion vector MV 1 is calculated as:

MVP_list1=((8−0)/(8−4))MV0=MV0<<1

The scaling of motion vector MV 0 is used such that the second motion vector predictor MVP_list 1 is shown as pointing to reference picture 504 - 2 . The difference between motion vector MV 1 and the second motion vector predictor MVP_list 1 can then be calculated and coded.

FIG. 8 shows a second example where motion vector MV 0 points to a past reference picture 504 - 1 and motion vector MV 1 points to a future reference picture 504 - 2 according to one embodiment. The MVP candidate for motion vector MV 1 is calculated as follows:

MVP_list1=((8−16)/8−4))MV0=−(MV0<<1)

As shown in FIG. 8 , current picture 502 has a POC of 8, reference picture 504 - 1 has a POC of 4, and reference picture 504 - 2 has a POC of 16. The sign is negative is this case to indicate the second MVP predictor MVP_list 1 is in the opposite direction of motion vector MV 0 .

Additionally, the other calculations described above with respect to the temporal difference can also be applied using POC. For example, motion vector predictors may be used to calculate second motion vector predictor MVP_list 1 .

FIG. 9 depicts a simplified flowchart 900 of a method for determining a second motion vector predictor according to one embodiment. At 902 , the method receives a current unit of video content. The current unit of video content is coded in a bi-prediction mode. At 904 , the method determines a motion vector predictor candidate set for a first motion vector for the current unit of video content. At 906 , a first motion vector predictor is determined from the motion vector prediction candidate set for the first motion vector. At 908 , the method calculates a second motion vector predictor for a second motion vector for the current unit of video content. The second motion vector predictor is calculated based on the first motion vector or the first motion vector predictor.

MVP Calculation for Two PUs within a CU

FIG. 10A and FIG. 10B depict an example of calculating first and second motion vector predictor MVP_list 0 or/and MVP_list 1 of the second PU based upon the information of the first PU where multiple PUs exist within a CU according to one embodiment. It will be understood that other partitioning of the CU may be appreciated. In FIGS. 10A and 10B , PU 0 is considered as the first PU in the CU and is followed by PU 1 in terms of coding order. These two PUs should have a high correlation and the MVPs for PU 0 and PU 1 may be similar.

Particular embodiments derive the MVP for PU 1 from the MVP of PU 0 . PU 0 and PU 1 may be any combination of uni-prediction and bi-prediction. For example, PU 0 and PU 1 may both be uni-prediction, PU 0 may be uni-prediction and PU 1 is bi-prediction, or PU 0 and PU 1 are both bi-prediction. In one example, PU 0 may have up to two motion vectors MV 0 _PU 0 and MV 1 _PU 0 that point to two reference pictures indexed by a first reference list ref_PU 0 _list 0 and a second reference list ref_PU 0 _list 1 . Similarly, PU 1 may have up to two motion vectors MV 0 _PU 1 and MV 1 _PU 1 that point to two reference pictures indexed by a first reference list ref_PU 1 _list 0 and a second reference list ref_PU 1 _list 1 . First motion vector predictors MVP 0 _PU 0 and MVP 1 _PU 0 for PU 0 are first determined, and then second motion vector predictors MVP 0 _PU 1 and MVP 1 _PU 1 for PU 1 can then be derived from first motion vector predictors MVP 0 _PU 0 and MVP 1 _PU 0 . The following calculation may scale first motion vector predictors MVP 0 _PU 0 and MVP 1 _PU 0 . For example, the scaling may be based on temporal distance or POC. The following may be used to calculate the first motion vector predictor MVP 0 _PU 1 for PU 1 :

MVP0_PU1=(TR0_PU1/TR0_PU0)*MVP0_PU0

Similarly, the second motion vector predictor MVP 1 _PU 1 for PU 1 is calculated from MVP 1 _PU 0 . A specific example is as follows:

MVP1_PU1=(TR1_PU1/TR1_PU0)*MVP1_PU0

In on embodiment, the above two equations, TR 0 _PU 0 is the difference between a current picture POC and a reference picture POC pointed to by motion vector MV 0 _PU 0 , TR 0 _PU 1 is the difference between the current picture POC and the reference picture POC pointed to by motion vector MV 0 _PU 1 , TR 1 _PU 0 is the difference between the current picture POC and the reference picture POC pointed to by MV 1 _PU 0 , and TR 1 _PU 1 is the difference between the current picture POC and the reference picture POC pointed to by motion vector MV 1 _PU 1 . Other temporal differences may also be used.

A PU may not use both reference list 0 and reference list 1 . For example, PU 0 may only use either reference list 0 or reference list 1 . On the other hand, PU 1 may still use both reference list 0 and reference list 1 . In this case, second motion vector predictors MVP 0 _PU 1 and MVP 1 _PU 1 can be derived from either first motion vector predictors MVP 0 _PU 0 or MVP 1 _PU 0 . For example, if only MVP 0 _PU 0 is available, second motion vector predictors MVP 0 _PU 1 and MVP 1 _PU 1 can be determined as follows:

MVP0_PU1=(TR0_PU1/TR0_PU0)*MVP0_PU0

MVP1_PU1=(TR1_PU1/TR0_PU0)*MVP0_PU0

In another case, if both PUs are in uni-prediction, if only MVP 0 _PU 0 is available, second motion vector predictor MVP 0 _PU 1 is determined from MVP 0 _PU 0 . FIG. 11 depicts a simplified flowchart 1100 of a method for determining MVPs when two PUs are found in a coding unit according to one embodiment. At 1102 , the method receives a first unit of video content. At 1104 , the method receives a second unit of video content. The first unit of video content (PU 0 ) and the second unit of video content (PU 1 ) are within a third unit of video content (CU). The first or second unit of video content may be coded in a uni-prediction or bi-prediction mode. This example is where both units are in bi-prediction, but other examples will be appreciated. At 1106 , a first motion vector predictor is determined for the first motion vector of the first unit of video content. At 1108 , a second motion vector predictor is determined for the second motion vector of the first unit of video content. At 1110 , the method calculates a first motion vector predictor for the second unit of video content from the first motion vector predictor of the first unit of video content and a second motion vector predictor for the second unit of video content from the second motion vector predictor of the first unit of video content.

Encoder and Decoder Examples

FIG. 12A depicts an example of encoder 402 according to one embodiment. A general operation of encoder 402 will now be described; however, it will be understood that variations on the encoding process described will be appreciated by a person skilled in the art based on the disclosure and teachings herein.

For a current PU, x, a prediction PU, x′, is obtained through either spatial prediction or temporal prediction. The prediction PU is then subtracted from the current PU, resulting in a residual PU, e. A spatial prediction block 1204 may include different spatial prediction directions per PU, such as horizontal, vertical, 45-degree diagonal, 135-degree diagonal, DC (flat averaging), and planar.

Temporal prediction block 1206 performs temporal prediction through a motion estimation operation. The motion estimation operation searches for a best match prediction for the current PU over reference pictures. The best match prediction is described by a motion vector (MV) and associated reference picture (refIdx). The motion vector and associated reference picture are included in the coded bit stream.

Transform block 1207 performs a transform operation with the residual PU, e. Transform block 1207 outputs the residual PU in a transform domain, E.

A quantizer 1208 then quantizes the transform coefficients of the residual PU, E. Quantizer 1208 converts the transform coefficients into a finite number of possible values. Entropy coding block 1210 entropy encodes the quantized coefficients, which results in final compression bits to be transmitted. Different entropy coding methods may be used, such as context-adaptive variable length coding (CAVLC) or context-adaptive binary arithmetic coding (CABAC).

Also, in a decoding process within encoder 402 , a de-quantizer 1212 de-quantizes the quantized transform coefficients of the residual PU. De-quantizer 1212 then outputs the de-quantized transform coefficients of the residual PU, E′. An inverse transform block 1214 receives the de-quantized transform coefficients, which are then inverse transformed resulting in a reconstructed residual PU, e′. The reconstructed PU, e′, is then added to the corresponding prediction, x′, either spatial or temporal, to form the new reconstructed PU, x″. A loop filter 1216 performs de-blocking on the reconstructed PU, x″, to reduce blocking artifacts. Additionally, loop filter 1216 may perform a sample adaptive offset process after the completion of the de-blocking filter process for the decoded picture, which compensates for a pixel value offset between reconstructed pixels and original pixels. Also, loop filter 1216 may perform adaptive loop filtering over the reconstructed PU, which minimizes coding distortion between the input and output pictures. Additionally, if the reconstructed pictures are reference pictures, the reference pictures are stored in a reference buffer 1218 for future temporal prediction.

FIG. 12B depicts an example of decoder 404 according to one embodiment. A general operation of decoder 401 will now be described; however, it will be understood that variations on the decoding process described will be appreciated by a person skilled in the art based on the disclosure and teachings herein. Decoder 404 receives input bits from encoder 400 for encoded video content.

An entropy decoding block 1230 performs entropy decoding on the input bitstream to generate quantized transform coefficients of a residual PU. A de-quantizer 1232 de-quantizes the quantized transform coefficients of the residual PU. De-quantizer 1232 then outputs the de-quantized transform coefficients of the residual PU, E′. An inverse transform block 1234 receives the de-quantized transform coefficients, which are then inverse transformed resulting in a reconstructed residual PU, e′.

The reconstructed PU, e′, is then added to the corresponding prediction, x′, either spatial or temporal, to form the new reconstructed PU, x″. A loop filter 1236 performs de-blocking on the reconstructed PU, x″, to reduce blocking artifacts. Additionally, loop filter 1236 may perform a sample adaptive offset process after the completion of the de-blocking filter process for the decoded picture, which compensates for a pixel value offset between reconstructed pixels and original pixels. Also, loop filter 1236 may perform adaptive loop filtering over the reconstructed PU, which minimizes coding distortion between the input and output pictures. Additionally, if the reconstructed pictures are reference pictures, the reference pictures are stored in a reference buffer 1238 for future temporal prediction.

The prediction PU, x′, is obtained through either spatial prediction or temporal prediction. A spatial prediction block 1240 may receive decoded spatial prediction directions per PU, such as horizontal, vertical, 45-degree diagonal, 135-degree diagonal, DC (flat averaging), and planar. The spatial prediction directions are used to determine the prediction PU, x′.

A temporal prediction block 1206 performs temporal prediction through a motion estimation operation. A decoded motion vector is used to determine the prediction PU, x′. Interpolation may be used in the motion estimation operation.

Parallel Processing of MVPs for PUs

In one embodiment, if there are two PUs within one CU, MVP manager 406 processes these two PUs in parallel. However, conventionally, the right PU in FIG. 14A or the bottom PU in FIG. 14B were not parallel processed because they have to use the left or the above PU in the same CU as one of spatial neighbors. Particular embodiments only use PUs in spatial neighboring CUs as spatial candidates for a current PU. This way, MVP manager 406 parallel processes PUs in the same CU.

FIGS. 15A and 15B depict an example of blocks to use as candidates according to one embodiment. In FIG. 15A , a block A 0 , instead of block A 1 (i.e., PU 0 ), is used as the left spatial block for the right PU (i.e., PU 1 ). Blocks A 2 and A 3 are still treated as the left-bottom and the left-above blocks for the right PU (PU 1 ). Similarly, in FIG. 15B , block B 0 , instead of block B 1 (i.e., PU 0 ), is used as the above spatial block for the bottom PU (i.e., PU 1 ), and blocks B 2 and B 3 are treated as the above-right and the above-left blocks for the bottom PU (PU 1 ). Thus, PU 0 is not used as a spatial neighboring block in the processing of PU 1 to find an MVP. This allows PU 0 and PU 1 to be parallel processed to find the motion vector predictor for PU 0 and PU 1 . In an alternative embodiment, MVP manager 406 treats the left neighbor in the example of FIG. 14A and the above neighbor in the example of FIG. 14B as not available. Thus, these two neighbors are not used in the MVP calculation and PU 0 and PU 1 can be parallel processed.

In order to improve the coding performance, particular embodiments may modify indexing of spatial and temporal neighbors. More specifically, if there is only one PU within one CU, the current indexing is kept unchanged, where the index for the left neighbor block is smaller than that for the above neighbor block.

If there are two PUs within one CU, the index for the left and the above block may be swapped. The swapping may depend on different factors, such as on PU partition types. For example, FIGS. 15A and 15B show different partition types according to one embodiment. In FIG. 15A , if the PU types are N×2N, nL×2N, or nR×2N, for the left PU (i.e., PU 0 ), the index for the left neighboring block (i.e., A 0 ) may be assigned a smaller number than that for the above neighboring block (i.e., A 3 ), and for the right PU (i.e., PU 1 ), the index for the above neighboring block (i.e., A 3 ) may be assigned a smaller number than that for the left neighboring block (i.e., A 0 ).

In FIG. 15B , if the PU types are 2N×N, 2N×nU or 2N×nD, for the above PU (i.e., PU 0 ), the index for the above neighboring block (i.e., B 0 ) may be assigned a smaller number than that for the left neighboring block (i.e., B 3 ), and for the bottom PU (i.e., PU 1 ), the index for the left neighboring block (i.e., B 3 ) may be assigned a smaller number than that for the above neighboring block (i.e., B 0 ).

Particular embodiments may be implemented in a non-transitory computer-readable storage medium for use by or in connection with the instruction execution system, apparatus, system, or machine. The computer-readable storage medium contains instructions for controlling a computer system to perform a method described by particular embodiments. The instructions, when executed by one or more computer processors, may be operable to perform that which is described in particular embodiments.

As used in the description herein and throughout the claims that follow, “a”, “an”, and “the” includes plural references unless the context clearly dictates otherwise. Also, as used in the description herein and throughout the claims that follow, the meaning of “in” includes “in” and “on” unless the context clearly dictates otherwise.

The above description illustrates various embodiments along with examples of how aspects of particular embodiments may be implemented. The above examples and embodiments should not be deemed to be the only embodiments, and are presented to illustrate the flexibility and advantages of particular embodiments as defined by the following claims. Based on the above disclosure and the following claims, other arrangements, embodiments, implementations and equivalents may be employed without departing from the scope hereof as defined by the claims.