International organisation for standardisation

6.8.3Motion/Mode Data Storage

6.8.3.1.1.1.1.1.1JCTVC-F472 Modified motion data storage reduction method [T. Sugio, T. Nishi (Panasonic)]

In this contribution, a modified motion data storage reduction method was proposed. It was proposed that a representative motion data was searched among 4x4 blocks according to encoding order. Experimental results reportedly showed 0.06% BR saving for HE and 0.04% BR savings for LC on average in the RA scenarios, and 0.1% BR saving for HE and 0.2% BR saving for LC on average relative to HM3.0+bugfix146.

Related to JCTVC-E231.

The proposal involves some extra work for the decoder. Because of this, and because the gain is rather small, stability of the current design seems like a higher priority.

6.8.3.1.1.1.1.1.2JCTVC-F482 Verification result of Qualcomm and NEC's Temporal QP Memory Compression (JCTVC-F499) [M. Shima (Canon)] [late upload 07-08]
6.8.3.1.1.1.1.1.3JCTVC-F060 Reducing Line Buffers for Motion Data and CABAC [T.-D. Chuang, C.-Y. Chen, Y.-W. Huang, S. Lei (MediaTek)]

Line buffers of 10.7 Kbytes are asserted to be required to store motion and CABAC context data of upper largest coding units (LCUs) for real-time decoding of 4Kx2K video in hardware when HM-3.0 is considered. In this contribution, for reducing motion line buffer size, a motion data compression method was proposed. For reducing CABAC line buffer size, CABAC context formation processes of several syntax elements were modified to avoid using upper block data and to only use left block data when the upper block belongs to the upper LCU. Simulation results reportedly showed that the motion data compression and the modified CABAC could achieve 56% and 74% reductions of motion and CABAC line buffers with 0.0% and 0.1% bit rate increases when the motion data compression ratio is 2:1 and 4:1, respectively. No encoding or decoding time increase was reported.

The initial suggested conclusion was that this is considered desirable, but should not be included yet at this stage of maturity – pending stabilization of other aspects such as spatial MV prediction.

Decision: The 2:1 variation was later included in the BoG recommendation JCTVC-F744, and was agreed to be adopted. However, during integration work after the meeting, it was noticed that the text and software submitted for the proposal actually also modified the boundary strength classification decisions made in the deblocking filter process, which could potentially have an impact on subjective quality. However, subjective evaluation was not done in Torino. It was then agreed (after the meeting) that rather than adopting this technique into the design at this time, it should be further studied in the CE12 Core Experiment on Deblocking Filtering to ensure that it does not have any adverse effect on the deblocking filter.

6.8.3.1.1.1.1.1.4JCTVC-F670 Cross-check of Mediatek's Reducing Line Buffers for Motion Data and CABAC (JCTVC-F060) [V. Sze (TI)] [late reg. 07-06, upload 07-14 after opening]