Back to Search Document details
36th Meeting: Kemer, TR, November 2024 2024-11-05 14:57
Non-EE2: Improved LFNST/NSPT kernel set selection for SGPM
Abstract
In ECM, SGPM mode uses the split direction to determine the transform kernel set for coding of the predicted residual signal. In this proposal, it is proposed to apply DIMD process to the prediction signal of the SGPM to compute the VIPM, and to determine the transform kernel set. We compare VIPM and the true intra modes in SGPM using a threshold.
JVET-AJ0350 Non-EE2: Improved LFNST/NSPT kernel set selection for SGPM [S. Puri, C. Bonnineau, K. Naser, F. Le Léannec (InterDigital)] [late]

In ECM, SGPM mode uses the split direction to determine the transform kernel set for coding of the predicted residual signal. In this proposal, it is proposed to apply DIMD process to the prediction signal of the SGPM to compute the VIPM, and to determine the transform kernel set. We compare VIPM and the true intra modes in SGPM using a threshold.

The following results are obtained compared to ECM-14.0:

AI {-0.03% Y, -0.01% U, -0.02% V, 100.0% EncT, 100.2% DecT}

Presented Tue 5 Nov. 1745 (chaired by JRO)

The proposal has relatively small gain, but it appears to be somewhat simpler in terms of additional processing than JVET-AJ0155, which has almost identical gain (but gains would probably not add up) with comparably low runtimes.

Several experts supported investigation of JVET-AJ0350 in an EE, as it straightforwardly re-uses some elements that are already used elsewhere.

Other (1)

Contributions in this area were discussed at 1750–1805 on Wednesday & Nov. 2024 (chaired by V. Seregin).

Decisions
Contributions in this area were discussed at 1750–1805 on Wednesday & Nov. 2024 (chaired by V. Seregin).
Citation