JVET-AF0012 JVET AHG report: Enhanced compression beyond VVC capability (AHG12) [M. Karczewicz, Y. Ye, L. Zhang (co-chairs), B. Bross, R. Chernyak, X. Li, K. Naser, Y. Yu (vice-chairs)]
Activities
The Common Test Conditions were updated (JVET-AE2017). The peak memory usage has been added to the reporting template. The primary activity of the AHG was the “Exploration experiment on enhanced compression beyond VVC capability” (JVET-AE2024). The combined improvements of the ECM-10.0 over VTM-11.0ecm10.0 anchor for AI, RA and LB configurations are:
All Intra Main 10 | |||||
Y | U | V | EncT | DecT | |
Class A1 | -10.55% | -21.81% | -29.24% | 915.8% | 439.3% |
Class A2 | -16.68% | -28.60% | -31.54% | 916.2% | 458.0% |
Class B | -11.02% | -26.58% | -24.82% | 826.4% | 455.0% |
Class C | -11.07% | -16.52% | -17.06% | 898.0% | 452.7% |
Class E | -14.59% | -24.40% | -22.89% | 804.1% | 467.8% |
Overall | -12.49% | -23.52% | -24.63% | 867.2% | 454.4% |
Class D | -9.17% | -14.38% | -14.35% | 851.0% | 445.1% |
Class F | -26.21% | -35.26% | -35.17% | 541.1% | 452.7% |
Class TGM | -39.28% | -47.12% | -46.45% | 457.3% | 496.2% |
Random Access Main 10 | |||||
Y | U | V | EncT | DecT | |
Class A1 | -22.25% | -27.01% | -36.07% | 882.3% | 784.9% |
Class A2 | -25.50% | -35.13% | -38.75% | 791.3% | 889.0% |
Class B | -20.50% | -34.42% | -32.13% | 673.3% | 731.6% |
Class C | -21.91% | -26.29% | -26.07% | 708.2% | 777.7% |
Class E | |||||
Overall | -22.23% | -30.91% | -32.63% | 744.0% | 784.1% |
Class D | -22.87% | -27.80% | -28.10% | 738.6% | 851.2% |
Class F | -28.30% | -36.71% | -37.10% | 602.2% | 482.5% |
Class TGM | -36.63% | -43.02% | -43.08% | 570.5% | 415.0% |
Low Delay B Main 10 | |||||
Y | U | V | EncT | DecT | |
Class A1 | |||||
Class A2 | |||||
Class B | -17.55% | -38.06% | -35.34% | 589.9% | 668.2% |
Class C | -19.29% | -28.96% | -29.31% | 648.1% | 710.0% |
Class E | -16.12% | -28.03% | -27.57% | 642.1% | 459.1% |
Overall | -17.77% | -32.52% | -31.39% | 621.7% | 620.8% |
Class D | -21.17% | -30.48% | -30.27% | 656.7% | 758.7% |
Class F | -24.60% | -37.20% | -36.64% | 588.5% | 498.3% |
Class TGM | -35.27% | -43.12% | -43.29% | 560.3% | 418.2% |
The rate reduction for camera-view sequences over VTM in RA configuration for {Y, U, V} increased from ECM-9.0’s {-21.03%, -28.42%, -29.93%} to ECM-10.0’s {-22.23%, -30.91%, -32.63%}. For SCC sequences (class TGM) the rate reduction for RA configuration increased from ECM-9.0’s {-35.17%, -41.66%, -41.35%} to ECM-10.0’s {-36.63%, -43.02%, -43.08%}.
Contributions
In addition to 24 EE2 contributions, 48 (comparing to 52 last meeting) EE2-related and AHG12-related contributions were received. There are also 7 contributions to AHG6 and AHG10 which address ECM encoder performance. The EE2-related and AHG12-related contributions can be subdivided as follows:
Intra (7)
- JVET-AF0059, "AHG12: Fix to interpolation filter for intra prediction", K. Andersson (Ericsson)
- JVET-AF0083, "Non-EE2: Enhancements on CCP merge for chroma intra coding", H. Huang, Y. Yu, H. Yu, D. Wang (OPPO), Z. Deng, K. Zhang, L. Zhang (Bytedance)
- JVET-AF0106, "EE2-related: Non-adjacent spatial candidates for DIMD merge", J. Huo, J. Fan, Z. Zhang, Y. Ma, F. Yang (Xidian Univ.), M. Li (OPPO)
- JVET-AF0114, "AHG12: Local-Boosting on Cross-Component Merge Mode", H. Qin, K. Ding, Z. Xu (TCL)
- JVET-AF0176, "Non-EE2: Decoder Derived Cross-Component Prediction", Y.-J. Chang, P.-H. Lin, V. Seregin, J.-L. Lin, M. Karczewicz (Qualcomm)
- JVET-AF0199, "Non-EE2: Bilateral Filtering for Intra Prediction", W. Yin, K. Zhang, Y. Wang, L. Zhang (Bytedance)
- JVET-AF0246, "AHG12: Harmonization of intra prediction mode derivation from neighbouring blocks", D. Kim, K. Kim, J.-H. Son, J. -S. Kwak (WILUS)
IntraTMP and IntraBC (15)
- JVET-AF0053, "EE2-related: Test 2.5a and TIMD fusion improvement", P. Andrivon, M. Blestel (Ofinno), K. Naser, F. Le Léannec, P. Bordes, F. Galpin, A. Robert (InterDigital)
- JVET-AF0072, "AHG12: FIBC flag inherits IntraTMP-FLM flag", M.-H. Jia, Y. Gao, Y. -X. Bai, S.-W. Xie, P. Wu, C. Huang (ZTE)
- JVET-AF0084, "Non-EE2: Update on IBC-LIC Model Merge mode", L. Zhang, Y. Yu, H. Yu, D. Wang (OPPO)
- JVET-AF0115, "Non-EE2: IBC with a further upward-extended reference area for screen content", Y. Kidani, H. Kato, K. Kawamura (KDDI)
- JVET-AF0116, "Non-EE2: slope adjustment for IBC LIC", C. Ma, X. Xiu, W. Chen, H.-H.Jhu, C.-W. Kuo, N. Yan, X. Wang (Kwai)
- JVET-AF0117, "Non-EE2: extensions of IBC GPM", C. Ma, X. Xiu, W. Chen, H.-H.Jhu, C.-W. Kuo, N. Yan, X. Wang (Kwai)
- JVET-AF0129, "Non-EE2: Auto relocated block vector prediction", N. Zhang, K. Zhang, L. Zhang (Bytedance)
- JVET-AF0132, "Non-EE2: Optimization of TIMD blending mode", C. Zhou, Z. Lv (vivo)
- JVET-AF0137, "AHG12: IntraTMP with Merge Candidates", K. Naser, F. Le Léannec, A. Robert, F. Galpin (InterDigital)
- JVET-AF0161, "AHG12: Harmonized IBC/ITMP search area with adaptive sampling", D. Ruiz Coll, B. Chen, (Ofinno), K. Naser, P. Bordes, F. Le Léannec, F. Galpin, (InterDigital)
- JVET-AF0162, "AHG12: Fixes to template matching", J. Chen, R.-L. Liao, X. Li, Y. Ye (Alibaba)
- JVET-AF0166, "Non-EE2: Intra TMP fusion probing", J.-L. Lin, P.-H. Lin, Y.-J. Chang, Z. Zhang, V. Seregin, M. Karczewicz (Qualcomm)
- JVET-AF0184, "Non-EE2: Search range optimization for Intra TMP", G. Verba, Z. Zhang, V. Seregin, M. Karczewicz (Qualcomm)
- JVET-AF0226, "AHG12: SGPM with IntraTMP and IBC", K. Naser, F. Le Léannec, P. Bordes, Y. Chen (InterDigital)
- JVET-AF0248, "Non-EE2: TIMD fusion optimization", P. Andrivon, M. Blestel (Ofinno), C. Zhou, Zhuoyi Lv (vivo)
Inter (12)
- JVET-AF0113, "Non-EE2: Modifications of affine merge candidates", K. Kim, D. Kim, J.-H. Son, J.-S. Kwak (WILUS)
- JVET-AF0118, "Non-EE2: Regression-based GPM blending", P. Bordes, F. Galpin, K. Naser, F. Urban, K. Reuze, F. Le Leannec, E. François (InterDigital)
- JVET-AF0121, "AHG12: Adjusting out-of-boundary prediction samples", P. Astola, J. Lainema (Nokia)
- JVET-AF0134, "AHG12: AMVP with SbTMVP mode", R.-L. Liao, J. Chen, Y. Ye, X. Li (Alibaba)
- JVET-AF0135, "AHG12: Additional SbTMVP candidates", R.-L. Liao, J. Chen, Y. Ye, X. Li (Alibaba)
- JVET-AF0160, "Non-EE2: On DMVR Extensions", M. Salehifar, Y. He, K. Zhang, L. Zhang (Bytedance)
- JVET-AF0168, "Non-EE2: Geometric partitioning mode with affine prediction", K. Zhang, Z. Deng, L. Zhang (Bytedance)
- JVET-AF0170, "Non-EE2: CIIP with subblock-based motion compensation", L. Zhao, K. Zhang, L. Zhang (Bytedance)
- JVET-AF0171, "Non-EE2: Sample distance-based weighting for CIIP", J. Zhao, S. Kim (LGE)
- JVET-AF0191, "Non-EE2: Enhancements on local illumination compensation", X. Xiu, C. Ma, N. Yan, H.-J. Jhu, C.-. Kuo, W. Chen, X. Wang (Kwai)
- JVET-AF0194, "Non-EE2: On LIC flag in merge mode", Y. Zhang, V. Seregin, H. Wang, Z. Zhang, C.-C. Chen, H. Huang, M. Karczewicz (Qualcomm)
- JVET-AF0200, "Non-EE2: Extension of local illumination compensation", Y. Wang, K. Zhang, Y. He, H. Liu, L. Zhang (Bytedance)
In Loop Filters (4)
- JVET-AF0169, "Non-EE2: Adaptive clipping with signalled lower and upper bounds", K. Cui, Z. Zhang, H. Huang, V. Seregin, M. Karczewicz (Qualcomm)
- JVET-AF0178, "Non-EE2: Applying fixed filters to output of the Gaussian filter", N. Hu, M. Karczewicz, V. Seregin (Qualcomm)
- JVET-AF0179, "Non-EE2: Fixed filter for Chroma ALF", N. Hu, M. Karczewicz, H. Wang, V. Seregin (Qualcomm)
- JVET-AF0198, "Non-EE2: Coefficient Precision Adjustment for ALF", W. Yin, K. Zhang, L. Zhang (Bytedance)
Transform (2)
- JVET-AF0082, "Non-EE2: On LFNST/NSPT for inter coding", F. Wang, J. Gan, Y. Yue, H. Yu, D. Wang (OPPO)
- JVET-AF0104, "Non-EE2: Direct Decoder-side Intra Mode Derivation for IntraTMP blocks", J. Huo, N. Qiu, Y. Ma, F. Yang (Xidian Univ.), Ming Li (OPPO)
Coefficients and Entropy Coding (4)
- JVET-AF0151, "EE2 related: CABAC parameters retraining", F. Galpin, F. Lo Bianco, C. Salmon-Legagneur, K. Naser (InterDigital)
- JVET-AF0164, "AHG12: CABAC inter/intra model switch", F. Lo Bianco, F. Galpin, C. Salmon-Legagneur (InterDigital)
- JVET-AF0185, "Non-EE2: Transform Coefficient Coding", P. Nikitin, M. Karczewicz, M. Coban, V. Seregin (Qualcomm)
- JVET-AF0202, "EE2 related: CABAC context initialization retraining and slice type based window offsets", V. Seregin, M. Karczewicz (Qualcomm)
Other (4)
- JVET-AF0058, "AHG12: GOP-based RPR encoder control for ECM", K. Andersson, J. Ström, R. Yu, W. Ahmad, P. Wennersten (Ericsson)
- JVET-AF0101, "AHG12: ECM software cleanup for decoder side intra predictions acceleration", Z. Fan, Y. Yasugi, T. Ikai (Sharp)
- JVET-AF0165, "Ahg12: On VTM-11ecm anchor low-delay test conditions", F. Le Léannec, S. Puri, E. François, K. Naser (InterDigital)
- JVET-AF0187, "Compression of Gaming Contents", Z. Lin, K. Cai, J. Sauer, Y. Zhao, E. Alshina (Huawei)
Recommendations
The AHG recommended:
- To review all the related contributions.