JVET-S0264 Performance Comparison of Screen Content Coding between HEVC and VVC [X. Xu, S. Liu (Tencent)] [late]
This was discussed during 1945–2000 on 26 June (chaired by JRO).
This document provides coding performance comparisons between HEVC and VVC, focusing on screen content materials. The simulation results using HEVC and VVC reference software (HM, SCM and VTM) are provided and compared. It is reported that VTM-9 software achieves around 61% BD rate reductions on 4:2:0 TGM class for 3 tested conditions as compared to HM-16.20; while the gains on 4:2:0 TGM 1080p class are 14%/27%/34% for AI/RA/LDB conditions when compared to SCM-8.6. The performance differences are roughly in line with 4:2:0 results when 4:4:4 materials are tested.
SCC performance comparisons between HEVC (HM and SCM) and VVC (VTM), 4:2:0 format
All Intra Main 10 | ||||||||||
Over HM16.20 | Over SCM-8.6 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
Class F | -39.39% | -39.90% | -42.49% | 4894% | 201% | -23.94% | -21.90% | -25.32% | 1627% | 190% |
Class TGM 1080p | -62.74% | -62.30% | -62.16% | 4043% | 176% | -14.41% | -12.85% | -12.61% | 1603% | 190% |
Random Access Main 10 | ||||||||||
Over HM16.20 | Over SCM-8.6 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
Class F | -41.55% | -44.78% | -46.09% | 653% | 196% | -29.86% | -31.94% | -33.51% | 630% | 164% |
Class TGM 1080p | -60.76% | -62.20% | -62.33% | 572% | 177% | -27.20% | -28.43% | -29.01% | 747% | 157% |
Low delay B Main 10 | ||||||||||
Over HM16.20 | Over SCM-8.6 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
Class F | -42.77% | -44.36% | -44.85% | 528% | 166% | -34.18% | -34.10% | -34.32% | 407% | 127% |
Class TGM 1080p | -61.48% | -63.18% | -62.72% | 451% | 159% | -34.19% | -35.29% | -34.85% | 468% | 131% |
SCC performance comparison between HEVC (HM) and VVC (VTM), 4:4:4 format
All Intra Main 10 | ||||||||||
YUV 4:4:4 | RGB 4:4:4 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
TGM 1080p | -68.12% | -68.80% | -69.03% | 4878% | 129% | -70.88% | -70.53% | -69.81% | 5415% | 117% |
TGM 720p | -51.78% | -54.35% | -57.11% | 4608% | 160% | -58.90% | -54.18% | -57.39% | 5257% | 147% |
Animation | -30.26% | -39.40% | -40.07% | 5948% | 206% | -40.17% | -38.02% | -33.30% | 7787% | 192% |
Mixed content | -50.17% | -54.23% | -54.42% | 5311% | 167% | -56.73% | -52.18% | -52.31% | 5826% | 156% |
Overall | -51.49% | -55.25% | -56.29% | 5100% | 160% | -57.84% | -54.96% | -54.69% | 5896% | 148% |
Random access Main 10 | ||||||||||
YUV 4:4:4 | RGB 4:4:4 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
TGM 1080p | -61.35% | -63.38% | -63.25% | 524% | 168% | -65.09% | -65.72% | -64.14% | 594% | 167% |
TGM 720p | -51.22% | -54.47% | -58.07% | 447% | 189% | -57.30% | -47.60% | -54.87% | 561% | 194% |
Animation | -34.41% | -41.84% | -42.78% | 733% | 186% | -44.96% | -40.32% | -32.93% | 954% | 200% |
Mixed content | -48.92% | -54.13% | -54.03% | 475% | 177% | -57.08% | -48.49% | -51.20% | 602% | 182% |
Overall | -50.02% | -54.24% | -55.41% | 527% | 180% | -56.83% | -51.41% | -52.03% | 649% | 185% |
Low delay B Main 10 | ||||||||||
YUV 4:4:4 | RGB 4:4:4 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
TGM 1080p | -56.74% | -59.05% | -58.93% | 324% | 153% | -60.77% | -61.67% | -60.01% | 353% | 148% |
TGM 720p | -49.38% | -53.31% | -56.98% | 271% | 144% | -54.75% | -44.86% | -51.36% | 326% | 150% |
Animation | -34.96% | -40.48% | -41.93% | 464% | 161% | -44.49% | -40.24% | -32.98% | 579% | 168% |
Mixed content | -47.31% | -53.82% | -54.39% | 287% | 152% | -57.52% | -48.78% | -49.89% | 361% | 155% |
Overall | -47.95% | -52.31% | -53.76% | 324% | 152% | -54.87% | -49.51% | -49.58% | 386% | 154% |
SCC performance comparison between HEVC (SCM) and VVC (VTM), 4:4:4 format
All Intra Main 10 | ||||||||||
YUV 4:4:4 | RGB 4:4:4 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
TGM 1080p | -12.24% | -7.09% | -7.28% | 1790% | 190% | -12.06% | -8.67% | -7.46% | 1985% | 181% |
TGM 720p | -16.89% | -11.99% | -11.51% | 1757% | 185% | -18.23% | -6.12% | -16.69% | 1919% | 174% |
Animation | -22.33% | -21.18% | -21.72% | 1912% | 196% | -21.85% | -17.21% | -15.90% | 2321% | 188% |
Mixed content | -17.33% | -12.49% | -13.14% | 1871% | 188% | -21.27% | -8.42% | -14.60% | 1918% | 178% |
Overall | -16.82% | -12.67% | -12.84% | 1823% | 189% | -17.89% | -9.72% | -13.43% | 2018% | 180% |
Random access Main 10 | ||||||||||
YUV 4:4:4 | RGB 4:4:4 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
TGM 1080p | -22.48% | -18.31% | -18.45% | 774% | 155% | -21.29% | -21.56% | -18.62% | 812% | 159% |
TGM 720p | -24.76% | -22.13% | -21.90% | 798% | 169% | -25.53% | -9.21% | -24.34% | 876% | 175% |
Animation | -29.70% | -29.96% | -31.11% | 634% | 166% | -31.67% | -24.05% | -20.03% | 756% | 180% |
Mixed content | -25.94% | -24.17% | -24.51% | 841% | 149% | -31.71% | -13.43% | -25.00% | 879% | 159% |
Overall | -25.42% | -23.15% | -23.45% | 762% | 160% | -26.96% | -16.82% | -21.92% | 831% | 168% |
Low delay B Main 10 | ||||||||||
YUV 4:4:4 | RGB 4:4:4 | |||||||||
Y | U | V | EncT | DecT | Y | U | V | EncT | DecT | |
TGM 1080p | -27.38% | -24.29% | -24.49% | 449% | 136% | -24.73% | -25.95% | -23.23% | 465% | 134% |
TGM 720p | -31.41% | -30.89% | -31.60% | 423% | 116% | -30.39% | -16.12% | -29.12% | 458% | 119% |
Animation | -30.67% | -30.23% | -31.22% | 397% | 137% | -30.07% | -24.89% | -19.84% | 457% | 149% |
Mixed content | -37.13% | -36.93% | -37.87% | 448% | 122% | -41.16% | -26.13% | -33.73% | 481% | 124% |
Overall | -31.32% | -30.16% | -30.83% | 430% | 127% | -31.02% | -22.95% | -26.44% | 465% | 130% |
The 4:2:0 test was without palette in VVC (as this is not in 4:2:0 profile), while SCM uses palette.
The SCM also used whole-frame IBC referencing, which is not supported in VVC.
These results indicated that VVC still achieves significant gain over the HEVC SCC profile. This is remarkable as the screen content specific tools are conceptually quite similar (except TSRC which is new), and even better designed for implementation such as small reference area for IBC. When IBC search range would be restricted in HEVC, the gap would probably be much larger.
No specific action was necessary on this.