Skip to content

JVET-AH0080 inference

Du Liu requested to merge duliu/vvc-software-vtm-nnvc:JVET_AH0080_inference into VTM-11.0_nnvc

This is the inference code of JVET-AH0080. Training code will be in a new MR. A small modification of SADL is needed, which will be in another MR.

I added a configuration parameter "NnlfTransInput" to both encoder configuration and decoder configuration, because currently HOP and LOP use the same inference code, and this is used to distinguish whether the model needs transformed inputs.

Currently, NnlfTransInput is set to true by default. Thus, for LOP.3, there is no need to specify NnlfTransInput. For HOP, "--NnlfTransInput=0" will be needed.

Merge request reports