Closed galoisgroupcn closed 5 years ago
Hello Mr.junyanz i am doing my master's research in I2I using GANs, i am planning to use cycleGAN as a baseline model but i am confused about the improvements because alot of work is done in this side like multi-model image2image translation which disentangle the content and style latent space, some use the idea of random noise, some used shared weights in encoder-decoder etc. do you have any suggestion for me like how to improve? thanks
These are all good ideas. I don't have a concrete idea on how to improve at this moment.
@junyanz Hello Dr. Zhu!
I'm quite impressed by your work. Do you think this repo can be successfully run on Google Colab?
Problem is, Colab goes with CUDA 10 (cannot downgrade) which does not work with Torch. There's a way to build Torch with CUDA 10 though, but must be hard to implement on Colab.
https://github.com/torch/cutorch/issues/834
-- Do you think Nvidia GTX 1050 (2GB RAM) can run this repo?
Look forward to your reply.
Thank you,
Amber Ji