about multi-gpu support

[pipipopo]

Hi, thanks for your amazing work. I have a problem with the multi-gpu support. The original batchsize is 2 for a real image and another random. In fact it's one pair. If we want to use multi-gpu for faster training, the batch size should be larger. But simply changing the batchsize causes error. May I know is it easy to modify the code for this purpose?

[junyanz]

What is your error? Have you set --gpu_ids?

[pipipopo]

RuntimeError: Expected tensor for argument #1 'input' to have the same device as tensor for argument #2 'weight'; but device 1 does not equal 0 (while checking arguments for cudnn_convolution)
The error is above when I use config below:
GPU_ID=0,1
--gpu_ids ${GPU_ID}
--batchSize 4

[pipipopo]

create web directory ../checkpoints/edges2shoes/edges2shoes_bicycle_gan/web...
Traceback (most recent call last):
File "./train.py", line 34, in
model.optimize_parameters()
File "../BicycleGAN-master/models/bicycle_gan_model.py", line 211, in optimize_parameters
self.update_G_and_E()
File "../BicycleGAN-master/models/bicycle_gan_model.py", line 199, in update_G_and_E
self.backward_EG()
File "../BicycleGAN-master/models/bicycle_gan_model.py", line 151, in backward_EG
self.loss_G_GAN = self.backward_G_GAN(self.fake_data_encoded, self.netD, self.opt.lambda_GAN)
File "../BicycleGAN-master/models/bicycle_gan_model.py", line 143, in backward_G_GAN
pred_fake = netD(fake)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 114, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 124, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 65, in parallel_apply
raise output
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 41, in _worker
output = module(*input, **kwargs)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "../BicycleGAN-master/models/networks.py", line 263, in forward
result.append(self.modeli)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/modules/container.py", line 91, in forward
input = module(input)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "../anaconda2/envs/xx/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 301, in forward
self.padding, self.dilation, self.groups)
RuntimeError: Expected tensor for argument #1 'input' to have the same device as tensor for argument #2 'weight'; but device 1 does not equal 0 (while checking arguments for cudnn_convolution)

[pipipopo]

After reading pytorch docs i still don't know how to set the data to devices in BicycleGAN....

[pipipopo]

Should
net = torch.nn.DataParallel(net, gpu_ids)
cooperate with any
data.to(device)?

[junyanz]

I found the issue. If you set --netD basic_256 and --netD2 basic_256, it should work. It seems that the multi-GPU issue comes from multi-scale D (--netD basic_256_multi). I haven't had time to address the issue there. I also updated the code so that it is compatible with pix2pix/CycleGAN. You may want to check out the latest commit.