International organisation for standardisation organisation internationale de normalisation

5.13.11HRD issues

No contributions particularly on this subject were noted (although HRD issues were noted to be mentioned in some other contributions).

5.14Quantization

See also notes in the CE4 section (section 4.4.1), especially relating to H0710 and H0730.

5.14.1QP prediction / delta QP coding

5.14.1.1.1.1.1.1.1JCTVC-H0189 Non-CE4: Low-complexity dQP coding [K Sato (Sony)] [late]

QP prediction has been investigated in CE4 subtest 1. The method proposed in JCTVC-G1024 appeared to be the best spatial QP prediction in terms of coding efficiency among the proposals in the previous CE4, and it is also asserted to be rate control friendly.

At the last meeting, some expert commented that he would preferr having a simpler method such as just averaging.

This document provides results with the QP prediction method of averaging the leftQP and topQP.

Compared to CE4 subtest 1.3: The same left/top averaging method as used for inter and intra; not using mode dependency any more, resulting in a gain that is 0.4% for intra, and 0.3/0.2% for RA/LD.

5.14.1.1.1.1.1.1.2JCTVC-H0204 Non-CE4: Compatible QP prediction with RC and AQ [H. Nakamura, M. Nishitani, S. Fukushima (JVC Kenwood)]

This contribution proposed a spatial QP prediction method under restricting QP references across LCU boundaries. The proposed method is based on the CE4 Subtest 1.3 and was modified to use the QP of the preceding "quantization group of coding units" (QGCU) more frequently for further rate control friendliness. In the combination with the QP prediction based on intra prediction direction in CE4 Subtest1.3, it was suggested to be expected to achieve higher efficiency improvement. The simulation results in the case of MaxCUDepthDQP=3(8x8) were reported as follows. The proposed method with the QP prediction based on intra prediction direction reportedly provided 0.5%, 0.4%, 0.3%, 0.3%, 0.2%, 0.3% and 0.3% luminance BD bit rate gains for AI HE, AI LC, RA HE, RA LC, RA HE10, LD HE, and LD LC configurations, respectively. The proposed method without the QP prediction based on intra prediction direction provided 0.3%, 0.3%, 0.3%, 0.3%, 0.2%, 0.3% and 0.3% luminance BD bit rate gains for AI HE, AI LC, RA HE, RA LC, RA HE10, LD HE, and LD LC configurations, respectively.

There is a difference compared to H0189: A non-available value (left or top) is replaced by the previous QP, and averaging is performed on whatever is remaining (i.e. if both are not available, the previous QP is averaged with itself). This scheme had slightly less rate reduction than H0189 (0.3% instead of 0.4%) for the AI case, and otherwise similar performance, but perhaps a bit more regular data flow.

Decision: Adopted.

5.14.1.1.1.1.1.1.3JCTVC-H0194 Non-CE4: Cross-verification report for modified spatial QP prediction based on intra prediction direction by JVC Kenwood (method 1 of JCTVC-H0204) [H. Aoki, K. Chono (NEC)]

5.14.1.1.1.1.1.1.4JCTVC-H0601 Non-CE4: Crosscheck report of Compatible QP prediction with RC and AQ (JCTVC-H0204) [M. Shima (Canon)] [late]
5.14.1.1.1.1.1.1.5JCTVC-H0289 Non-CE4: Spatial QP prediction based on motion vector predictors (AMVP) [P. Andrivon, P. Bordes (Technicolor)]

This contribution presented results of spatial QP prediction based on:

• 
  Test 1: motion vector predictor cues (in Inter),
  
• 
  Test 2: a combination of test 1 and an intra direction based method (JCTVC-G1204 Subtest 1.3),
  
• 
  Test 3: a non-mode-dependent QP prediction (above and left QGCU averaging if within LCU otherwise last coded QP).
  

Simulations results reportedly showed on average 0.2% to 0.5% BD bit rate gains for test 1 and test 2 in RA, LB and AI cases, respectively, under the test condition defined in CE4 Subtest 1 (AVC like).

Compared to the non-mode-dependent QP prediction (test 3), the gains are equivalent.

Test 3 is same as H0189.

5.14.1.1.1.1.1.1.6JCTVC-H0525 Non-CE4: Cross-verification report for spatial QP prediction based on motion vector predictors (JCTVC-H0289) [H. Aoki, K. Chono (NEC)] [late]

5.14.1.1.1.1.1.1.7JCTVC-H0188 On the syntax of ChromaQpOffset [K Sato (Sony)]

At the 7^th JCT-VC meeting, ChromaQpOffset, the same as the one in AVC, was adopted. This document proposed two modifications:

• 
  Moving ChromaQpOffset from PPS to APS
  
• 
  Changing the names as "Cb_QP_Offset" and "Cr_QP_Offset"
  

5.14.1.1.1.1.1.1.8JCTVC-H0400 Chroma QP extension [J. Xu (Sony), A. Tourapis (Magnum), E. Maani, K. Sato (Sony)]

In HM5.0, Chroma QP can only take values in the range of [0, 39]. This contribution proposed to extend the range of Chroma QP values to [0, 51]. The proponent asserted that this change enhances the functionality, flexibility and friendliness of HEVC for various potential applications, such as rate control at low bit-rates, including the flexibility to obtain the desired performance balance point between luma/chroma, and for handling video sources with different colour formats (e.g. RGB). The contribution proposed the formula QPCx = Clip(0, 51, QPY + Cx_QP_offset).

The initial Track B recommendation was to adopt the chroma QP mapping from H0400, change the names as suggested in H0188, and keep the offset values in PPS (as these seem not to need frequent switching). This was further discussed in a plenary discussion, and further study was recommended with no immediate action.

This, as proposed, could be a significant change to the balance between luma and chroma in our common test conditions. No subjective testing had been done. In the reported experiment results, the balance between luma and chroma fidelity was changed substantially. It was remarked that the non-linear mapping has been used in standards for avoiding colour artefacts at low bit rates for a long time. It was suggested to be questionable whether the suggested default value is appropriate. Further study was suggested to identify whether this really adds the desired flexibility for rate control without having impact on subjective quality.