JVET-AF0296 AhG11: on HOP learning rate [F. Galpin, T. Dumas, P. Bordes (InterDigital)] [late]

A Unified Filter Architecture for the High-performance Operation Point (HOP) was defined at JVET-AD0380. It includes filter architecture, training process and training test set. Training strategy and full results were presented in JVET-AF0041. This contribution gives additional results on the impact of the learning rate on the training results. It is reported that a larger learning rate improves the results. Results compared to original HOP in VTM mode are {} and {} for AI and RA configurations respectively.

See notes for JVET-AF0150. Qualcomm was willing to help.

Other aspects of neural network-based video coding (0)

Section kept as a template for future use.

AHG6/AHG12: Enhanced compression beyond VVC capability (78+1)

Summary and BoG reports (1)

Contributions in this area were discussed at 1400–1915 on Friday 13 Oct. 2023 (chaired by JRO) except otherwise noted.