JVET-AM0195 AhG18: Gradual Decoding Refresh (GDR) under ultra low latency test scenario [M. Sychev, A. Dzugaev, S. Ikonin, E. Alshina (Huawei)]
This informational proposal provides the results and analysis of Gradual Decoding Refresh technology under ultra-low latency (ULL) test scenario investigated by AhG18. The BD-PSNR results demonstrate a performance of this technique under ULL test conditions: 1.63 dB increase on average for classes A & B at same latency restrictions. Freezing time is decreased by 10% on average for classes A & B. As a supplementary result by request during AhG18 discussions the GDR combined with Scalable (SGDR) solution was evaluated demonstrating synergy effect of efficiency. SGDR further improves the quality under ULL scenario up to 4.44 dB of average performance increase in respect to single layer VTM and freezing time was decreased by 26%. Demo videos solution are provided in the same conditions.
Decision(SW/BF): Include bug fix in GDR encoder to AHG branch (not relevant for main VTM branch).