Code for Minibatch Discrimination on MNIST Missing

[wenyangfu]

Hi, I've been trying to reproduce a minibatch discrimination GAN for MNIST based on the paper, but I keep getting poor results. Would it be possible for "train_mnist_minibatch_discrimination.py" to be uploaded to the repo? I assume it exists, since MNIST digits generated via minibatch discrimination were shown in the paper. Thanks for your time!

[xunhuang1995]

+1

[djsutherland]

:+1: We're trying to compare to your results, and it'd be really helpful to be able to be sure we didn't get something stupid wrong in plugging together train_cifar_minibatch_discrimination and train_mnist_feature_matching.

[djsutherland]

@wenyangfu @xunhuang1995: I talked to the authors about this offline, and they told me that the just used the same model as for CIFAR/SVHN for MNIST minibatch. My fork has a train_mnist_minibatch_discrimination.py that implements that (just loading the MNIST data instead of CIFAR, basically).

[rafaelvalle]

https://github.com/openai/improved-gan/blob/master/mnist_svhn_cifar10/nn.py

[AilsaF]

Hi @dougalsutherland, just wondering does it mean minibatch can only work with conv discriminative model instead of dense layer one?

[djsutherland]

@AilsaF I don't know if it wouldn't work with dense layers, just that they used the convolutional one. You could try it. :)