search

Netflix / vmaf

Perceptual video quality assessment based on multi-method fusion.
Other
4.59k stars 749 forks source link

The well-trained models of SpatioTemporal VMAF and Ensemble VMAF - release? #497

Closed Sissuire closed 4 years ago

Sissuire commented 4 years ago

Here it is. I just read the newly published literature SpatioTemporal Feature Integration and Model Fusion for Full Reference Video Quality Assessment, Christos G. Bampis, Zhi Li and Alan C. Bovik, IEEE TCSVT, 2018.

I had checked the source files of ST-RRED, SpEED, or ST-MAD, but not found the corresponding models for ST-VMAF and E-VMAF. That is, I can only extract features with these sources but fail to predict the final video quality as the literature says.

Do the models have been released? waiting for a demo as usage in consideration of its quite promising performance.

christosbampis commented 4 years ago

Hi, we have not published the ST-VMAF model yet; for E-VMAF you can rely on the python class EnsembleVmafQualityRunner to get a feeling of how to apply the models on a new test video. One would have to train two models separately and then plug them in to the aforementioned class.