Joint Collaborative Team on Video Coding (jct-vc) Contribution

5.6MV prediction and coding

JCTVC-B023 [J. Lim, S. Park, B.-M. Jeon (LG)] Extended merging scheme using motion-hypothesis prediction

This document presented an extended merging process for inter prediction. In the merging process of TMuC, the current motion partition is predicted with one of two neighboring partitions, above and left partitions. In the proposed method, the merging process is combined with motion hypothesis prediction. When enabled, the current partition is predicted with the weighted sum of the two candidate partitions. One additional flag is signaled when condition is satisfied (merge_flag equal to 1, both neighboring partitions are available, and motion parameters of the two candidate partitions are different). The purpose of the proposed method is to enhance the coding efficiency of P picture by extending the current merging process. The simulation result shows that the proposed method achieves 2.9% and 3.3% bit rate reduction under Hier-P and IPPP condition on average for class B, C and D sequences. Moreover, the proposed method reportedly introduced negligible complexity increment relative to TMuC with the merging process.

Remark: Can achieve the same functionality by coding the picture as a B picture. Comparisons should perhaps be relative to what a B picture can accomplish. This aspect should be studied.

JCTVC-B061 [Y. Su, A. Segall (Sharp Labs)] Verification of JCTVC-B023 (Extended merging scheme using motion-hypothesis prediction)

This contribution reports verification results of JCTVC-B023 "Extended merging scheme using motion-hypothesis prediction" from LG Electronics. The verification package contains 1) bitstreams of the proposed scheme, 2) decoder executable of the proposed scheme and 3) spreadsheets for verification. The verification is carried out by 1) decoding the bitstreams, 2) collect PSNR/bitrate numbers and 3) verify that the numbers generated match those in the spreadsheets.

In summary, verification of JVTVC-B023 succeeded on the following test cases:

HierP on classB (100frames) sequences, classC sequences and classD sequences

IPPP on classB (100frames) sequences, classC sequences and classD sequences

In the accompanied spreadsheets, the verified numbers are shaded in blue color.

Again, executable is not our preferred method of verification.

JCTVC-B071 [T. Yoshino, S. Naito, S. Sakazawa (KDDI)] Refinement of skip coding mode for HEVC

As the inter coding scheme in AVC, the "skip" coding mode is effective as well as Inter16x16 mode especially under low bit rate condition. As the result of macroblock (MB) basis coding mode decision based on R-D optimization criteria, those two modes are frequently selected under the low bit rate condition because the R-D optimal point locates between those two candidates on the approximated R-D curve.

This contribution introduces an additional coding mode for P-picture based on extending conventional skip mode in order to improve coding performance for high resolution video. From experimental result, it was reported that an average bit rate reduction for IpPpP coding of 2.2% and 3.6% was achieved for the coding of all available sequences and for the coding of the HD sequences (1080p and 720p), respectively.

Remark: Seems like just an entropy coding alternative representation of zero motion vector delta with residual. Can analyze this as an entropy coding difference, and perhaps as a different way to decide what to do.

Further study was encouraged to determine where this gain is coming from, since that is not clear at present.

JCTVC-B072 [Y.-W. Chen, C.-H. Wu, C.-L. Lee, T.-W. Wang, W.-H. Peng (NCTU/ITRI)] MB Mode with joint application of template and block motion compensations

This contribution introduced a MB mode, termed the P+ mode, to improve the prediction efficiency of P-MBs by a joint application of template and block motion compensations. The motion vector (MV) found by minimizing template matching error is viewed as an additional free MV, which can contribute to estimating pixel intensities in a P-MB. As in POBMC (of JCTVC-A123), the predictors derived from the template and block MVs are linearly combined based on a distance weighting criterion. In particular, given that the template MV tends to minimize the prediction error in the upper left quarter of a MB, the block MV search criterion is changed so that the resulting MV can contribute more to minimizing the error in the remaining part. When applied to 16x16 P-MBs, the proposed scheme is observed to have an average BD-Rate saving of 2.4% and a corresponding BD-PSNR gain of 0.1 dB. It achieves the highest rate saving (4.2%) in S03 (Kimono, Class B) and the lowest (0.4%) in S10 (PartyScene, Class C).

Sort of another way to introduce biprediction for a P picture. Should compare to B picture. Similar issue to JCTVC-B023 in terms of analysis.

The contributor indicated that they did not have time to test with B pictures.

JCTVC-B081 [K. Sato (Sony)] Proposals for improved MV coding

Motion vector competition scheme was proposed in VCEG-AC06 and has been implemented into the KTA software. Most responses to the HEVC CfP employ this scheme.

For further improvement this contribution proposed a dynamic code number assignment for motion vector predictors. The proposed method reportedly also improves subjective quality between the moving object and still background in fixed-camera-view sequences.

Test results are needed – but are not yet available.

Question: Within a given test sequence, how often would the selection change?

Question: Benefit of set order selection that is at sequence level in JCTVC-A124 to be moved to slice/picture level.

JCTVC-B094 [F. Bossen, P. Kosse (DOCOMO USA Labs)] Simplified motion vector coding method

A method for coding motion vector data was presented. It was based on the method previously proposed in document JCTVC-A114. It reportedly achieves coding efficiency similar to the methods included in the TMuC software while providing reduced complexity by simplification of the parsing of motion vector data and reducing the number of motion vector predictors.

It seems like there is an opportunity for complexity reduction and for elimination of duplicate functionality.

JCTVC-B108 [E. Maani, W. Liu, A. Tabatabai (Sony)] Multi-parameter motion (MPM)

In this document, a motion compensation that uses of multiple prediction blocks was proposed – similar in concept to Flierl/Wiegand/Girod multi-hypothesis.

Have only tested with two MVs – pointing to same reference picture.

Similar comment to others: should compare to B, not P.

JCTVC-B101 [G. Clare, J. Jung, S. Pateux (Orange Labs)] Preliminary results on motion vector prediction

Was presented Thursday a.m. as evaluation of TMuC behavior.