JVET-AL0015 JVET AHG report: Gaming content compression (AHG15) [S. Puri, J. Sauer (co-chairs), R. Chernyak, A. Duenas, L. Wang, V. Zakharchenko (vice chairs)]
Software developed by AHG15 consists of forks of VTM/ECM that have support for reading auxiliary data that is provided for sequences of classes G1/G3. A third repository contains scripts for the verification of the auxiliary information. The repositories can be found at https://vcgit.hhi.fraunhofer.de/jvet-ahg-gcc.
- Test results
Test results were obtained according to JVET-AJ2027. The list of sequences/classes is copied for convenience below:
Class | Sequence name | Frame count | Low delay frame count | Frame rate | Bit depth | Intra | Random access | Low-delay |
G1 | Level1_SDR | 600 | 300 | 60 | 10 | M | M | M |
Darktree_SDR | 600 | 300 | 60 | 10 | M | M | M | |
ARPG2_SDR | 600 | 300 | 60 | 10 | M | M | M | |
DesertTown3_SDR | 600 | 300 | 60 | 10 | M | M | M | |
SunTemple3_SDR | 600 | 300 | 60 | 10 | M | M | M | |
G2 | GTAV | 600 | 300 | 60 | 8 | M | M | M |
Minecraft | 600 | 300 | 60 | 8 | M | M | M | |
JianlingTemple | 600 | 300 | 60 | 8 | M | M | M | |
BaoleiYard | 300 | - | 60 | 8 | M | M | - | |
G3 | Level1_HDR | 600 | 300 | 60 | 10 | O | O | O |
Darktree_HDR | 600 | 300 | 60 | 10 | O | O | O | |
ARPG2_HDR | 600 | 300 | 60 | 10 | O | O | O | |
DesertTown3_HDR | 600 | 300 | 60 | 10 | O | O | O | |
SunTemple3_HDR | 600 | 300 | 60 | 10 | O | O | O |
For class G3 the HDR versions of Level1 and Darktree are still missing.
VTM-11.0ecm16.0 vs ECM-16.0
SDR
All Intra Main 10 | ||||||
Over Anchor | ||||||
Y | U | V | EncT | DecT | VmPeak | |
Class G1 | -11.59% | -23.99% | -18.80% | 1153.56% | 617.35% | #REF! |
Class G2 | -18.38% | -22.87% | -27.01% | 1216.37% | 694.86% | #REF! |
Overall | -14.61% | -23.49% | -22.45% | 1181.07% | 650.67% | #REF! |
Random Access Main 10 | ||||||
Over Anchor | ||||||
Y | U | V | EncT | DecT | VmPeak | |
Class G1 | -27.49% | -42.71% | -34.63% | 921.82% | 1545.85% | #REF! |
Class G2* | -24.45% | -36.52% | -37.83% | 907.19% | 1517.11% | #REF! |
Overall | -26.35% | -40.39% | -35.83% | 916.31% | 1535.01% | #REF! |
* ECM crashed for BaoleiYard. Results are averaged without BaoleiYard
Low delay B Main 10 | ||||||
Over Anchor | ||||||
Y | U | V | EncT | DecT | VmPeak | |
Class G1 | -20.75% | -51.68% | -41.07% | 869.00% | 1318.26% | #REF! |
Class G2 | -18.03% | -47.02% | -46.82% | 907.92% | 1380.22% | #REF! |
Overall | -19.73% | -49.94% | -43.22% | 879.94% | 1335.67% | #REF! |
Low delay P Main 10 | ||||||
Over Anchor | ||||||
Y | U | V | EncT | DecT | VmPeak | |
Class G1 | -19.03% | -57.76% | -49.24% | 807.46% | 1077.85% | #REF! |
Class G2 | -15.86% | -54.20% | -54.39% | 831.03% | 1183.64% | #REF! |
Overall | -17.84% | -56.43% | -51.17% | 816.22% | 1116.37% | #REF! |
HDR
There are still no HDR versions for Level1 and Darktree.
All Intra | ||||||||||
Over VTM11.0ecm16.0 | ||||||||||
wPSNR | PSNR | |||||||||
DE100 | PSNR-L100 | Y | U | V | Y | U | V | EncT | DecT | |
Class G3 | -24.71% | -12.14% | -11.47% | -43.01% | -39.88% | -11.45% | -43.19% | -39.59% | 1475% | 592% |
Random Access | ||||||||||
Over VTM11.0ecm16.0 | ||||||||||
wPSNR | PSNR | |||||||||
DE100 | PSNR-L100 | Y | U | V | Y | U | V | EncT | DecT | |
Class G3 | -40.68% | -26.92% | -28.06% | -58.09% | -56.31% | -28.28% | -58.58% | -56.50% | 1018% | 1413% |
Low Delay B | ||||||||||
Over VTM11.0ecm16.0 | ||||||||||
wPSNR | PSNR | |||||||||
DE100 | PSNR-L100 | Y | U | V | Y | U | V | EncT | DecT | |
Class G3 | -36.74% | -20.79% | -19.98% | -61.91% | -57.96% | -20.26% | -62.59% | -57.91% | 823% | 1132% |
Low Delay P | ||||||||||
Over VTM11.0ecm16.0 | ||||||||||
wPSNR | PSNR | |||||||||
DE100 | PSNR-L100 | Y | U | V | Y | U | V | EncT | DecT | |
Class G3 | -36.46% | -20.41% | -19.68% | -61.93% | -56.74% | -19.94% | -62.57% | -56.64% | 953% | 1028% |
- Input contributions
- AHG9/AHG15: Depth-adaptive picture scaling information SEI (JVET-AL0212)
- Recommendations
The AHG recommended to:
- Review input documents on gaming content compression
- Identify if/how auxiliary information can be used for coding of gaming content