tensors are on different GPUs

[DragonBornHD]

I run the demo, return 'tensors are on different GPSs' , but I have only one GPU.
The demo was run successfully with CPU.
Can you put the process of using the GPU to release it?
Thank you very much!

[alexkoltun]

Most likely that the net is not on GPU, try adding the following, just after weight loading:

if torch.cuda.is_available():
    net = net.cuda()

[DragonBornHD]

I tried this way, another error occurred:

TypeError: mul received an invalid combination of arguments - got (torch.FloatTensor), but expected one of:

• (float value)
  didn't match because some of the arguments have invalid types: (torch.FloatTensor)
• (torch.cuda.FloatTensor other)
  didn't match because some of the arguments have invalid types: (torch.FloatTensor)

I added the 'net = net.cuda()' after 'net.load_state_dict(torch.load(trained_model))' , and 'x = x.cuda()' after 'x = Variable(x.unsqueeze(0))' .
Thank you.

[alexkoltun]

I see.. I think setting the default tensor to be cuda might solve it, try to put the following somewhere before model loading:

if torch.cuda.is_available():
    torch.set_default_tensor_type('torch.cuda.FloatTensor')

[DragonBornHD]

That's doesn't work, same problem.

[alexkoltun]

Did you restart the kernel?

I tested with the following change on a GPU machine and it works:

if torch.cuda.is_available():
    torch.set_default_tensor_type('torch.cuda.FloatTensor')
net = build_ssd('test', 300, 21)    # initialize SSD
net.load_weights('../weights/ssd300_mAP_77.43_v2.pth')
if torch.cuda.is_available():
    net = net.cuda()

[DragonBornHD]

Yes it works！
torch.set_default_tensor_type('torch.cuda.FloatTensor') Should before net = build_ssd('test', 300, 21)
Thank you very much !!

[DonghoonPark12]

@alexkoltun
I'm sorry but, when I train my network on gpu by below code,

if torch.cuda.is_available():
    net = net.cuda()

Why there is another need when I'm training for example in train.py

        if args.cuda:
            images = images.cuda()
            targets = [ann.cuda() for ann in targets]
        else:
            images = images
            targets = [ann for ann in targets]

Is there also diffenrence between below two line??

images = images.cuda()
# and
images = images

Thanks ahead.