Back to Search Document details
26th Meeting: by teleconference, April 2022 2022-04-22 01:03
EE1-1.2: Neural network based in-loop filter with a single model
Authors: Liqiang Wang Tencent Xiaozhong Xu Tencent Shan Liu Tencent Franck Galpin InterDigital
Abstract
This contribution reports the EE results of JVET-Y0080[]. A neural network based in-loop filter with constrained memory size and lower complexity is proposed in this contribution, compared with the previous implementation. In all, only one single model is used in the proposed filter design. In addition to the reconstruction image, other side information, such as the prediction image, slice QP, based QP and slice type, is fed into the network. Based on the EE1 anchor, the related results are shown in order of RA, LB and AI configurations as follows.
References:
JVET-W0113 JVET-Y0080
Cited:
JVET-AA0023 JVET-AA0088 JVET-AA0122 JVET-AA0131 JVET-AA0178 X JVET-AA0185 X JVET-AA0217 X JVET-Z0011 JVET-Z0178 X JVET-Z0234 JVET-Z2023
PATENTS:
US20230368077A1 0.70 2023-11-16 CN118632003A 0.34 2024-09-10 CN116976456A 0.32 2023-10-31 AU2022201052B2 0.32 2025-01-23 EP2840140B2 0.30 2023-02-22 AU2024213183A1 0.28 2024-10-03 CN113647119B 0.26 2024-07-02 US11282088B1 0.24 2022-03-22
ARCHIVE v4 :
JVET-Z0091-EE1-1.2.1.xlsm JVET-Z0091-EE1-1.2.2.xlsm JVET-Z0091-EE1-1.2.3.xlsm JVET-Z0091-v4.docx JVET-Z0091.pdf
Citation