JVET-AM0007 JVET AHG report: ECM tool assessment (AHG7) [X. Li (chair), L.-F. Chen, Z. Deng, J. Gan, E. François, R. Ishimoto, H.-J. Jhu, J. Lainema, X. Li, J. Pardo, A. Stein, H. Wang (vice chairs)]
Group off tests
Test settings and crosschecking
The following five groups of tools were defined.
- Group 1: Inter template matching tools
- Group 2: Coding tools that interleave the (merge/skip/AMVP/subblock/IBC/etc) list derivation with the intra prediction/reconstruction process
- Group 3: Intra and IBC template matching (with search) related tools
- Group 4: Tools that require more processing on the neighbouring reconstructed samples than VVC
- Group 5: Tools with large memory access
The planned testers and crosscheckers are identified in the table below.
Tests | Testers / Crosscheckers | Testers / Crosscheckers |
Group 1 off | Johan Pardo (johan.esprit.pardo1@huawei.com) | Charles Salmon-Legagneur (charles.salmon-legagneur@interdigital.com) |
Group 2 off | Jonathan Gan | Xinwei Li |
Group 3 off | Ryo Ishimoto (ishimoto.ryo@mail.sharp) | Zhipin Deng (zhipin.deng@bytedance.com) |
Group 4 off | Hong-Jheng Jhu | Xiang Li |
Group 5 off | Lien-Fei Chen (lienfei.chen@global.tencent.com) | Jani Lainema (jani.lainema@nokia.com) |
Group 1-5 off | Xiang Li | Hongtao Wang |
ECM-17.0 was used in the AHG7 tool off tests. The cfg files used are included in the ECM software package.
Group 1 off
Group 1 includes inter template matching tools. The offgroup1.cfg was used in addition to ECM CTC settings.
Group 2 off
Group 2 includes coding tools that interleave the (merge/skip/AMVP/subblock/IBC/etc) list derivation with the intra prediction/reconstruction process. The offgroup2.cfg was used in addition to ECM CTC settings.
Group 3 off
Group 3 includes intra and IBC template matching (with search) related tools. The offgroup3.cfg was used in addition to ECM CTC settings.
Group 4 off
Group 4 includes tools that require more processing on the neighbouring reconstructed samples than VVC.
The offgroup4.cfg file was used in addition to ECM CTC settings.
Group 5 off
Group 5 includes tools which need large memory access.
The offgroup5.cfg file was used in addition to ECM CTC settings.
Group 1-5 off
In this test, all the tools in the group 1-5 are switched off. The offgroup1-5.cfg was used in addition to ECM CTC settings.
Summary
The tool-off results on top of the recent ECM versions are summarized below for BD-PSNR-Y. Note that SCC results are not included.
Issues noted
Resolved issues
- #109, Current master tip (7fee7e00d) has decoding mismatch with AHG7 group1-5 off
- #107, Current master tip (06c21da0) decoding mismatch with AHG7 group2 off
- #106, Current master tip (06c21da0) fails MD5sum check at decoding with AHG7 group1 off
- #105, Current master tip (06c21da0) has encoding crash with AHG7 group1 off
- #104, Current master tip (d3a1301a) has encoding crash with AHG7 group1-5 off
Open issues
- #91, MaxTU setting
- #87, ECM-15.0 decoder crashes when "InterLFNST=0"
- #80, Compilation Issue ECM14.0 when disabling GPM reordering
- #78, Encoder crash in RA with --LMChroma=0
- #77, Encoder crash in RA with --DepQuant=0 --NumSignPred=0
- #71, Encode/decode mismatch when using single tree
- #66, Difference in encoding results of ECM-12.0 in AhG7 group 2 tool off test
- #65, Different encoding results of ECM12 in AHG7 group1-4 off tests
- #64, Encode/decode mismatch and decoder crash when inter-CCCM is disable
- #53, Decoding mismatch was observed when AMVR is off
Input contributions
2025-04-22 18:52:59 | 2025-04-22 19:28:04 | 2025-05-20 02:43:45 | Report of AHG7 conference call on Assessment Perspectives of Codec/Coding Tools | |||
2025-05-20 18:41:53 | 2025-05-20 18:45:18 | 2025-05-20 18:45:18 | AHG7: Summary on the Tool Analysis in Earlier Proposals/Reports | |||
2025-06-19 11:56:43 | 2025-06-19 14:19:37 | 2025-06-19 14:19:37 | AHG7: ECM tool combination | |||
2025-06-19 17:50:35 | 2025-06-19 17:56:52 | 2025-06-19 17:56:52 | On Analysing and Reporting Encoding and Decoding Complexity | |||
2025-06-19 22:17:19 | 2025-06-19 23:23:08 | 2025-06-19 23:23:08 | AHG7: TMRL Tool-Off Bug Fix | |||
2025-06-19 22:53:02 | 2025-06-23 20:11:35 | 2025-06-23 20:11:35 | AhG7: On bin to bit ratio in ECM | T. N. Canh, P. Yin, S. McCarthy (Dolby), J. N. Shingala (Ittiam) | ||
2025-06-25 06:18:58 | 2025-06-25 06:25:43 | 2025-06-25 06:25:43 | [AHG7] External memory bandwidth evaluation | Y. Kim, L. Li, M. W. Park, M. Park, M. Budagavi, K. P. Choi (Samsung) |
Recommendations
The AHG recommended the following:
- Continue and improve tool assessment
- Resolve identified software issues related to the tool assessment
- Review all the input documents
It was commented that newly integrated tools should report a mandatory tool-off test together with the merge request (this could be more relevant for future developments rather than ECM).
A BoG was planned for later in the meeting on methodology for complexity assessment of tools.
It was suggested to investigate whether the case groups 1-5 off could be combined with ECM5x used in CfE dry-run, to potentially generate another lower-complexity ECM version in the CfE part on faster encoding.