Ramana Sundararaman
Ramana Sundararaman
> If the idea was to have a reference for the minibatch discrimination layer, then yes. My gists only show how to train GANs based on the DCGAN paper, they...
How to test the Weight normalization ? Is it fine if i have a numpy implementation, normalizing rows and columns as two different functions and compare with what the layer...
Hi, I have made a draft here, https://github.com/Theano/Theano/pull/5189 . Could you please have a look and let me know if the Op is implemented correctly? Ramana
Hi, If anyone is interested, the code for `fast-RCNN`, with installation instructions are here, https://github.com/Lasagne/Recipes/issues/35#issuecomment-286567589
@f0k sorry about the delay, seems like this week, all the time that I said I'd be having are during the weekend. To my eyes, this PR is done. Would...
Updated grammar and fixed the Travis error. In case anything else needs to be done here, I'll fix it
I can dedicate 6-10~hrs from next week. @f0k I'll start one by one. Could you please point me out to issues based on relevance and it's urgency? Also, is there...
One comment is yet to be addressed(moving the assignment of `self.broadcastable` to the constructor, I'll push that commit by tonight.
Very sorry for the late reply. > Ensuring the correct broadcast pattern in MergeLayer So, we do the check which we have previously discussed, by iterating through the dimensions of...
> We could call self.output_shape at the end of every constructor to ensure the shapes are sane This seems to be an excellent idea to me. > because at that...