Dmitry Ulyanov

Hmm, do you have 32 bit version ?

Have no idea to be true..

Hello, find all `:cuda()` calls and remove them. Convert net from `cudnn` to `cunn` with `cudnn.convert(net, nn)`. Should be it!

You do not need GPU to convert. On Fri, Mar 3, 2017 at 8:35 PM Sachin Kelkar wrote: > To run cudnn.convert(net, nn), I need a GPU right? > I...

Hey, it's very simple, just replace all `:cuda()` to `:cl()` with find and replace :) I do not have time now to update the repo.

@zhreshold I also have this error turning up at random. And sometimes there is no error, but one of the frames in the end of the sequence is completely black....

But why the algorithms work so bad on batch15 image then? Did you test them on the full images, not on the chunks from this repo?

Hi, we have used a classifier, pretrained on the Imagenet. Note that inception score with such classifier does not make sense. We initially computed such value to compare to other...

Take a look at the log, is variance for real of fake images 0?

Take a look at a previous epochs, probably it is not 0 there. Otherwise there are some things you can do to get rid of collapsing. Try either reducing `z`...