Back to Search Document details
19th Meeting: by teleconference, June 2020 2020-06-16 00:10
AHG9: On ACT and LFNST
Authors: Mohammed Golam Sarwer Alibaba Y. Ye Alibaba Mohammed Sarwer
Abstract
The current VVC design allows cascaded invocation of three transforms such as ACT, primary transform and LFNST for a given coding unit. However, it is asserted that the coding gain of LFNST is significantly reduced when ACT is enabled, compared to when ACT is disabled. Since cascading three transforms incurs higher hardware cost (more clock cycles and increased latency of the decoding pipeline) for limited benefit, this contribution proposes to disallow ACT and LFNST combination. Given the late stage of VVC development, such disabling mechanism considers methods that requires only high level changes.
Meeting Notes
JVET-S0197 AHG9: On ACT and LFNST [M. G. Sarwer, Y. Ye (Alibaba)]
Cited:
JVET-S0011 JVET-S0144
PATENTS:
WO/2021/252645 0.91 20210092408 0.90 WO/2021/055781 0.90 20220329861 0.90 4087246 0.90 20210385491 0.70 WO/2021/216303 0.31 2023303717 0.26
ARCHIVE v2 :
JVET-S0197-Table-3-ACT-LFNST.xlsm JVET-S0197-Table-4-ACT-LFNST.xlsm JVET-S0197-Table-5-ACT-LFNST.xlsm JVET-S0197-Table-6-ACT-LFNST.xlsm JVET-S0197-Table-7-ACT-LFNST.xlsm JVET-S0197-v1.docx JVET-S0197-v2-clean.docx JVET-S0197-v2.docx
Decisions
JVET-S0197 AHG9: On ACT and LFNST [M. G. Sarwer, Y. Ye (Alibaba)]
Citation