Open neubig opened 6 years ago
just a memo:
a
conv_2d_tranpose
(i.e. deconv_2d) is needed in DyNet to build a dcgan example.
I am working on this issue.
@gpengzhi you're building a conv_2d_transpose
operator for dynet?
@pw2393 no, I am working on the example instead of the operator.
If you want to implement DCGAN, a conv2d_transpose op might be required.
Part of: https://github.com/clab/dynet/issues/1284
It'd be nice to have a competitive GAN example. One example is the PyTorch DCGAN:
We might also consider a GAN for text example, such as something related to style transfer.