JVET-AM0218 AHG18: Solution beyond scalable coding for ultra-low latency and packet loss resilience [S. Ikonin, V. Khamidullin, I. Gribushin, M. Sychev, X. Ma, E. Alshina (Huawei)]

This contribution describes the coding technology beyond scalable coding that can be used for ultra-low latency and packet loss resilient decoding scenarios. The described technology re-uses concept of data partitioning of H.263/4 which separates prediction and residual information into different sub-streams. Additionally, entropy coding of residual coefficients is performed using multiple independent entropy coded chunks. A lost aware NN-based loop filter is further applied to reduce subjective impact of lost residuals and propagation if these errors via inter frame prediction. A demo video files are provided with contribution.

Compared to and on top of a 2-layer SNR scalability with priority of base layer from JVET-AK0183 (vs. adding high priority DP part of the stream of the proposed approach, and the loss-aware loop filter). Gain is 2.2 dB on average.

About 3% overhead for the multiple chunks.

The loss aware loop filter gets information which parts of the picture are lost. It was trained on VTM streams using BVI-DVC plus error patterns.

Contribution for information – no specific action proposed.