JVET-L0009 JVET AHG report: Neural Networks in Video Coding (AHG9) [S. Liu, B. Choi, K. Kawamura, Y. Li, L. Wang, P. Wu, H. Yang]

This document summarizes the activity of AHG9: Neural network in video coding between the 11th meeting Ljubljana, SI (10–18 July 2018) and the 12th meeting in Macao, CN (3 – 12 Oct 2018).

The AHG used the main JVET reflector, jvet@lists.rwth-aachen.de, with [AHG9] in message headers. There was no email exchange on the main reflector and some offline discussions among proponents, participants and outside JVET. Academia universities and labs continued showing interests in the subject of Neural Networks for video compression with questions such as complexity and practicability, etc.

Input documents (technical proposals) related to AHG9 were identified as:

• JVET-L0242 “AHG9: Dense Residual Convolutional Neural Network based In-Loop Filter”, [Y. Wang, Z. Chen, Y. Li (Wuhan Univ.), L. Zhao (Tencent)]
• JVET-L0383 “AHG9: Convolution Neural Network Filter” [K. Kawamura, Y. Kidani, S. Naito (KDDI)]

The AHG recommended:

• To review all related contributions
• To continue discussions about methodologies and measurements for evaluating neural network related video coding tools

In the discussion, it was suggested that software availability would be helpful, including the tools for training. It was commented that it would not be feasible to have a CE until there is software.