JVET-T0041 Methodology and reporting template for neural network coding tool testing [S. Liu, A. Segall, E. Alshina, J. Boyce, M. Wien, D. Grois]
This document describes methodology for neural network-based video coding tool testing and provides a template for reporting the results of neural network-based video coding tools. It is an update of previous JVET output document JVET-M1006, which was used for reporting and evaluating proposed neural network-based video coding tools in related AHGs and Core Experiments. It is encouraged to use the provided template for reporting and analyzing the proposed neural network-based video coding tools in future JVET meetings. This was reviewed in the AHG meeting and was used for most of the incoming contributions.
This contribution was discussed at 0745 on Thursday 8 October (chaired by GJS & JRO).
It was agreed to produce a corresponding output after considering experiences with it.
Data was used from Univ. Bristol, which was posted publicly by them.
There was some discussion of soliciting comments from hardware experts about what complexity and gain target could be acceptable.
BoG work was suggested to come up with a group suggestion. See JVET-T0130 for the BoG report.