set weights to 1

[assafdv]

Hi,
can you explain why you set the gmm weights to 1 (FisherVector.py, line 170). it seems that the likelihood_ratio param should hold the posterior probabilities p(C|X) (dented as gamma in Perronnin et al.). For my understanding this is what you get when you call perdict_proba.

thanks,
Assaf.

[qinchuanhui]

I agree with that, the averaged weights (with 1) may get much worse result.