Why thresholding?

[josedvq]

Hi, thanks for the cool implementation. I am just wondering what the idea is of this line:

fisher_vectors[fisher_vectors < 10**-4] = 0

I see in the paper "Image Classification with the Fisher Vector: Theory and Practice" that they say they use a threshold (with the same value) but on the likelihood ratio, not on the final fisher vector, and I can't see a reason to threshold it like that. Thanks for your time!