search

zhangjunh / DR-GAN-by-pytorch

An implement of Disentangled Representation Learning GAN for Pose-Invariant Face Recognition
126 stars 37 forks source link

training on multi-gpu #11

Closed harshit4084 closed 5 years ago

harshit4084 commented 5 years ago

hello, I want to run this code on multi-gpu, can you tell what minor or major changes I need to do if you have done so earlier?

zhangjunh commented 5 years ago

Hi @harshit4084 , this project was finished last year and I didn't use multi-gpu when I code it. I'm not sure how to run this code on multi-gpu, so I couldn't help you. If I have the time in the future I will add it to the code, and now you might make it on your own.

harshit4084 commented 5 years ago

Hi @zhangjunh Yes. I converted the code for multi-gpu. And it is working fine for single DR GAN and giving some dim error when splitting on multiple gpus in Multi DRGAN. I'll put it here, you'll be able to interpret it better. Also, Approximately how much time did it take to train the model to give a decent output?

zhangjunh commented 5 years ago

Hi @harshit4084 , I couldn't remember the exact time for training munti-DRGAN. It's might be more than a week using single 1080TI.

MmDawN commented 3 years ago

Hi @zhangjunh Yes. I converted the code for multi-gpu. And it is working fine for single DR GAN and giving some dim error when splitting on multiple gpus in Multi DRGAN. I'll put it here, you'll be able to interpret it better. Also, Approximately how much time did it take to train the model to give a decent output?

Hello, I am also trying to use multiple GPUs to run this code, but after torch.nn.DataParallel processing , the batchsize of the three variables that input to the Encoder is inconsistent. Have you ever encountered this problem? If possible, can you provide us with the multi-GPU running solution you have implemented, thank you!