Organisation internationale de normalisation

5.9MVC test sequences

5.9.1.1.1JVT-Z035 / M15102 ( Info) [I. Feldmann, M. Mueller, F. Zilly, R. Tanger, K. Mueller, A. Smolic, P. Kauff, T. Wiegand (HHI)] Progress report on 3DTV video acquisition

This contribution contained a progress report about work on content creation for 3DTV. The document described a multiview camera arrangement, details of the camera hardware, planned content, disparity range, and potential use of the data.
This contribution had been reviewed by MPEG as M15102 and was not presented in detail to the JVT.

6New AVC Proposals

6.1Adaptive MV precision

6.1.1.1.1JVT-Z022 / M15185 ( Prop 2.2/3.1) [S. Sekiguchi, K.Otoi, Y. Yamada, K. Asai, T. Murakami (MEI)] 4:4:4 video coding perf with adaptive MV coding

This contribution reports a potential compression performance improvement of 4:4:4 video coding with adaptive coded representation of motion vector information depending on the magnitude of the motion vector. The objective of this study was to investigate the possibility of further compression performance improvement for a 4:4:4 profile in order to apply it to consumer-level applications requiring much higher compression ratios. In the case of high-compression conditions, it was suggested to be necessary to consider further reduction of the coded bits for motion information. The studied approach here is to perform an adaptive motion vector search that limits fractional sample accuracy depending on the magnitude of motion vector, and to derive a coded representation of motion vectors assuming the adaptive accuracy motion search. Experimental results reportedly showed an advantage to the studied approach in the case of high-compression conditions.
This was a contribution toward a hypothetical future 4:4:4 profile oriented toward consumer applications.
The contribution suggested to use a relatively small search range with 1/4-sample search, a larger range with 1/2-sample search, and a yet larger integer search range.
On one test sequence ("Shimoda"), compared to the current JM search and syntax technology, an experiment yielded a reported 5.6% delta bit rate improvement over the 38-42 dB fidelity range.
What does this have to do with 4:4:4? Couldn't it apply equally well to other chroma formats? Response: Yes, it can.
A 50-63% bit rate savings was reported for the motion vector data portion of the bitstream with IPPP coding.
The contribution asserts a savings in memory bandwidth also.
Remark: The entropy coding aspect seems difficult to comprehend, since motion vectors are sent as deltas from predictors, whereas the precision depends on the actual MV value result rather than the delta – how would we determine the interpretation of the MVD? Response: A scheme is used which seems to work; it involves maintaining quarter-sample precision for the predictor – other details were considered difficult to explain, but were reported to work.
Question: What happens if you change the Lambda for the anchor? Response: Don't know.
Question: Was this using CAVLC or CABAC? Answer: CAVLC.
Presentation? Uploaded later.
JVT disposition: Further investigation was encouraged, including testing on more data and for other chroma formats.