JVET-T0012 JVET AHG report: High bit depth, high bit rate, and high frame rate coding (AHG12) [A. Browne, T. Ikai, X. Xiu]
This AHG report was reviewed in session 5 at 0500 UTC Thursday 8 October (chaired by GJS & JRO).
This document summarizes the activity of AHG12: High bit depth, high bit rate, and high frame rate coding between the 19th meeting by teleconference (22 June – 1 July 2020) and the 20th meeting by teleconference (7–16 October 2020).
In total there were 11 contributions noted to be related to high bit depth and one on high bit rate, but none related to high frame rate. The contributions are listed in the following section.
- Test conditions and sequences
- JVET-T0047, AHG12: Preliminary common test conditions for high bit depth video coding, T. Ikai, A.Browne, X. Xiu
- JVET-T0082, New HLG sequences for high bit-depth video coding, K. Kondo, M. Ikeda, A. Browne (Sony)
- Tools and transforms
- JVET-T0084, AHG12: Transform coefficients range extension for high bit-depth coding, T. Zhou, T. Chujoh, E. Sasaki, T. Ikai (Sharp)
- JVET-T0086, AHG12: SIMD implementation of BDOF for high bit-depth coding, T. Chujoh, E. Sasaki, T. Ikai (Sharp)
- JVET-T0087, AHG12: VVC coding tool evaluation for high bit-depth coding, T. Ikai, T. Zhou, T. Hashimoto (Sharp)
- JVET-T0091, AHG12: Fix on encoder overflow issue of the LMCS at high bit-depth coding, X. Xiu, H.-J. Jhu, Y.-W. Chen, T.-C. Ma, X. Wang (Kwai)
- JVET-T0093, AHG12: On signalling for high-bit depth, Y. Kidani, K. Kawamura, K. Unno (KDDI)
- Golomb-Rice coding
- JVET-T0072, AHG12: Rice parameter selection for high bit depths, A. Browne, S. Keating, K. Sharman (Sony)
- JVET-T0085, AHG12: Rice parameter derivation for high bit-depth coding, T. Hashimoto, T. Ikai (Sharp)
- JVET-T0089, AHG12: Slice based Rice parameter selection for transform skip residual coding, H.-J. Jhu, X. Xiu, Y.-W. Chen, T.-C. Ma, C.-W. Kuo, X. Wang (Kwai Inc.)
- JVET-T0105, AHG12: On the Rice parameter derivation for high bit-depth coding, L. Pham Van, D. Rusanovskyy, M. Karczewicz (Qualcomm)
- Lossless coding (RGB and YCgCo)
- JVET-T0111 YCgCo-R: Observations and findings [D. Buitenhuis (VideoLAN), A.M. Tourapis (Apple Inc)]
The following results comparing the performance of VTM 10.0 and HM 16.22 were obtained using the preliminary test conditions (JVET-T0047). The preliminary test conditions include SVT content previously included in the HEVC RExt test conditions, and new 12 bit PQ content converted from the same original sources as the PQ content in the VVC 10 bit HDR test conditions.
Extended Precision Enabled
The following results were obtained with extended precision enabled (--ExtendedPrecision=1) in both VTM 10.0 and HM16.22. Note that the VVC specification does not have this flag, although the VTM supports it.
AI-12 | |||||
Over HM16.22 EP1 | |||||
Y | U | V | EncT | DecT | |
SVT | 0.01% | 0.24% | 0.14% | 12387% | 132% |
PQ444 | −4.78% | −6.16% | −6.84% | 4460% | 182% |
PQ422 | −8.43% | −12.76% | −13.44% | 4460% | 182% |
| |||||
Overall | −4.40% | −6.23% | −6.71% | 6040% | 162% |
LDB-12 | |||||
Over HM16.22 EP1 | |||||
Y | U | V | EncT | DecT | |
SVT | −3.59% | −1.19% | −1.19% | 1464% | 173% |
PQ444 | −7.18% | −8.58% | −8.23% | 1055% | 182% |
PQ422 | −9.47% | −14.41% | −15.61% | 1055% | 182% |
| |||||
Overall | −6.75% | −8.06% | −8.35% | 1298% | 180% |
RA-12 | |||||
Over HM16.22 EP1 | |||||
Y | U | V | EncT | DecT | |
SVT | −3.68% | −1.42% | −1.43% | 2414% | 206% |
PQ444 | −7.86% | −9.43% | −9.21% | 1512% | 200% |
PQ422 | −10.26% | −15.29% | −16.84% | 1512% | 200% |
| |||||
Overall | −7.27% | −8.71% | −9.16% | 1903% | 202% |
AI-16 | |||||
Over HM16.22 EP1 | |||||
Y | U | V | EncT | DecT | |
SVT | 36.73% | 34.40% | 34.33% | 11495% | 145% |
| |||||
Overall | 36.73% | 34.40% | 34.33% | 11495% | 145% |
LDB-16 | |||||
Over HM16.22 EP1 | |||||
Y | U | V | EncT | DecT | |
SVT | 19.93% | 21.07% | 21.13% | 1604% | 219% |
| |||||
Overall | 19.93% | 21.07% | 21.13% | 1604% | 219% |
RA-16 | |||||
Over HM16.22 EP1 | |||||
Y | U | V | EncT | DecT | |
SVT | 17.74% | 19.96% | 19.94% | 2494% | 219% |
| |||||
Overall | 17.74% | 19.96% | 19.94% | 2494% | 219% |
Extended Precision Disabled
The following results were obtained with extended precision disabled (--ExtendedPrecision=0) in both VTM 10.0 and HM16.22.
AI-12 | |||||
Over HM-16.22 | |||||
Y | U | V | EncT | DecT | |
SVT | 0.55% | 0.55% | 0.43% | 11780% | 144% |
PQ444 | −5.09% | −6.23% | −6.86% | 4414% | 191% |
PQ422 | −8.33% | −12.95% | −13.56% | 4414% | 191% |
| |||||
Overall | −4.29% | −6.21% | −6.66% | 5886% | 172% |
LDB-12 | |||||
Over HM-16.22 | |||||
Y | U | V | EncT | DecT | |
SVT | −3.09% | −0.82% | −0.80% | 1228% | 145% |
PQ444 | −7.84% | −8.88% | −8.50% | 1038% | 191% |
PQ422 | −9.72% | −14.96% | −15.99% | 1038% | 191% |
| |||||
Overall | −6.88% | −8.22% | −8.43% | 1212% | 175% |
RA-12 | |||||
Over HM-16.22 | |||||
Y | U | V | EncT | DecT | |
SVT | −3.20% | −1.16% | −1.15% | 2030% | 148% |
PQ444 | −8.35% | −9.75% | −9.40% | 1456% | 192% |
PQ422 | −10.51% | −15.72% | −17.03% | 1456% | 192% |
| |||||
Overall | −7.35% | −8.88% | −9.19% | 1759% | 177% |
AI-16 | |||||
Over HM-16.22 | |||||
Y | U | V | EncT | DecT | |
SVT | 29.80% | 36.03% | 34.31% | 7995% | 162% |
| |||||
Overall | 29.80% | 36.03% | 34.31% | 7995% | 162% |
LDB-16 | |||||
Over HM-16.22 | |||||
Y | U | V | EncT | DecT | |
SVT | 21.90% | 23.63% | 23.80% | 922% | 166% |
| |||||
Overall | 21.90% | 23.63% | 23.80% | 922% | 166% |
RA-16 | |||||
Over HM-16.22 | |||||
Y | U | V | EncT | DecT | |
SVT | 19.66% | 21.56% | 21.69% | 1482% | 167% |
| |||||
Overall | 19.66% | 21.56% | 21.69% | 1482% | 167% |
The results were reported to show that currently VTM 10.0 offers gains over HM 16.22 for most 12 bit processing except SVT AI. However, there are significant losses for 16 bit content. This performance was not unexpected, and some of the contributions listed above directly address this loss.
The AHG recommended the following:
- To review all related contributions
- To continue studying the benefits and characteristics of VVC coding tools for high bit depth and high bit rate coding.
- To continue investigating the requirements for future extensions of VVC to support high bit depth and high bit rate coding
- To continue refining the preliminary test conditions for high bit depth and high bit rate coding including the evaluation of new and modified test sequences
- To identify new test sequences for high frame rate coding which might be included in future test conditions
It was commented that for up to 12 bit, the extended precision may not be necessary. However, another participant said that some gain is shown for using this for some content.
It was commented that for the 16 bit SVT content, in some cases the extended precision flag is not helping. Another participant commented that without the extended precision flag enabled, the design basically does not support a full 16 bits of content, so the results on such content should be viewed skeptically.
See JVET-T0111 providing information about lossless coding, which reports that HEVC SCC is outperforming VVC for intra coding in some cases.