search

xh-liu / CC-FPSE

Code for NeurIPS 2019 paper "Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis"
128 stars 15 forks source link

How can I train on a single GPU? #7

Closed Ha0Tang closed 4 years ago

xh-liu commented 4 years ago

You can train on a single GPU by deleting the codes about multiple gpus and distributed multiprocessing. But the training speed would be very slow.

Ha0Tang commented 4 years ago

Thanks, how can I train the model on a DGX1?

Rubiel1 commented 4 years ago

Can you give us a sample script? how would a 1 gpu, pix2pix model with pix2pix data set look like?

Ha0Tang commented 4 years ago

did you solve it @Rubiel1 ?

Ha0Tang commented 4 years ago

@Rubiel1 Do you know how to train on 8 GPUs?

Ha0Tang commented 4 years ago

Solved.

Mut1nyJD commented 4 years ago

Can you give us a sample script? how would a 1 gpu, pix2pix model with pix2pix data set look like?

You need to change a bit. I had to remove the whole distributed code and also make sure all SyncBatchnorms gets replaced with normal BatchNorms most of this you can do via argument options.