JVET-P0970 AHG13: Additional ISP Tool-off tests [S. De-Luxán-Hernández, V. George, G. Venugopal, J. Brandenburg, B. Bross, H. Schwarz, D. Marpe, T. Wiegand (HHI)]

The contributor indicated that this document did not need detailed presentation. It is available for study, and is summarized below.

This document provides additional tool-off tests of the ISP coding mode with the LFNST combination proposed in JVET-P0392, for higher QP ranges than for CTC and for lossless coding. The reported losses when turning ISP off are:

• AHG13: AI: 0.41% (Y), 86% (Enc), 97% (Dec) RA: 0.20% (Y), 98% (Enc), 100% (Dec)
• JVET-P0392 t2: AI: 0.52% (Y), 85% (Enc), 98% (Dec) RA: 0.30% (Y), 96% (Enc), 99% (Dec)
• QP 27,32,37,42: AI: 0.60% (Y), 87% (Enc), 98% (Dec) RA: 0.27% (Y), 98% (Enc), 99% (Dec)
• QP 32,37,42,47: AI: 0.84% (Y), 89% (Enc), 98% (Dec) RA: 0.35% (Y), 98% (Enc), 100% (Dec)
• AHG18 lossless: AI: 0.67%, 84% (Enc), 106% (Dec) RA: 0.12%, 98% (Enc), 100% (Dec)

Neural networks (1)

Contributions in this category were discussed Wednesday 8 1750–1810 in Track A (chaired by JRO).