Back to Search Document details
30th Meeting: Antalya, TR, April 2023 2023-04-20 16:27
EE1-1.7: Optimization of training and network for NNVC filter set 0
Authors: Renjie Chang Tencent Liqiang Wang Tencent Xiaozhong Xu Tencent Shan Liu Tencent
Abstract
This contribution reports the EE results of JVET-AC0197. The proposed single-model filter in JVET-AC0197 is fine-tuned based on different training scripts for a better performance. Besides, a new combination network of resblock from filter set 0 and transformer block is designed to further improve the coding gains. Based on NNVC-4.0, the test results are shown in order of RA and AI configurations as follows.
Meeting Notes
JVET-AD0166 EE1-1.7: Optimization of training and network for NNVC filter set 0 [R. Chang, L. Wang, X. Xu, S. Liu (Tencent)]
Cited:
JVET-AD0329 X
ARCHIVE :
JVET-AD0166-Test-a_vs_Filter0.xlsm JVET-AD0166-Test-a_vs_NNVC-4.0.xlsm JVET-AD0166-Test-b_vs_Filter0.xlsm JVET-AD0166-Test-b_vs_NNVC-4.0.xlsm JVET-AD0166-Test-c_vs_Filter0.xlsm JVET-AD0166-Test-c_vs_NNVC-4.0.xlsm JVET-AD0166-v3.docx ~$ET-AD0166-v1.docx
Decisions
JVET-AD0166 EE1-1.7: Optimization of training and network for NNVC filter set 0 [R. Chang, L. Wang, X. Xu, S. Liu (Tencent)]
Citation