JVET-AL0201 AHG18: Performance evaluation of VTM under proposed ULL test conditions and proposed software modifications [S. Ikonin, V. Khamidullin, R. Shabaev, I. Gribushin, M. Sychev, E. Alshina (Huawei)]

Document JVET-AL0176 proposes methodology and test conditions for ultra-low latency (ULL) and error resilience performance evaluation in context of AHG18 work. This contribution describes proposed framework for ULL metrics evaluation. The exemplary quality-against-latency-restriction curves are provided for unicast and multicast scenario. Simulation results of VTM for single layer and scalable two-layer coding are provided. The contribution contains a list of modifications and fixes to enable AHG18 simulation software working according to proposed test conditions.

For the feedback scenario, losses are dependent on the latency. Decoder is performing frame freezing or generates pictures based on broken reference, and PSNR is measured against that.

Freeze ratio is also measured.

The decoder software might also be useful as basis for implementing more sophisticated concealment strategies.

As the frozen frame does not necessary look bad, PSNR may not be an appropriate metric. The same might be true if other concealment methods would be used which are not necessarily provide pixel fidelity.

Improvement is shown when using a scalable solution. However, in lossless case, bitrate is increased.It was commented that a very complex network model may not be needed for this study.

It was agreed to add the software to the AHG branch fur further study and basis for further developments.