ap229997 / Conditional-Batch-Norm

Pytorch implementation of NIPS 2017 paper "Modulating early visual processing by language"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

gammas and betas

0zgur0 opened this issue · comments

I think this variable should be 1-dimensional, gammas, and betas should be same over the batch, right?

self.betas = nn.Parameter(torch.zeros(self.batch_size, self.channels).cuda())

Yes, as per equation 2 in the paper Modulating early visual processing by language, it varies only channel wise. Thanks for pointing out this error.