What should I do with the weight type

1

Hi, I’m using the gdl loss function from https://github.com/ginobilinie/medSynthesisV1/blob/master/nnBuildUnits.py, when I call and run the code on my own, it shows the problem:

RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same
I’m new in pytroch, and have no idea what’s wrong with anywhere and what should I do for the next step. Any help will be highly appreciated!!

I call it in this way:

from model.unet2 import UNet
device = torch.device(“cuda” if torch.cuda.is_available() else “cpu”)
net = UNet(in_channels=1, out_channels=1, init_features=32)
net.to(device)
ctimgs= read_img(“k1/train/ct_data/patient_002_ct.gipl”)
mrimgs = read_img(k1/train/mri_data/patient_002_mr_T1.gipl")
ctimgs1 = ctimgs[0,:,:]
mrimgs1 = mrimgs[0,:,:]
ctimgs1 = ctimgs1[np.newaxis,np.newaxis,:,:]
mrimgs1 = mrimgs1[np.newaxis,np.newaxis,:,:]
ctimgs1 = torch.from_numpy(ctimgs1)
mrimgs1 = torch.from_numpy(mrimgs1)
out = net(mrimgs1)
gdlloss = gdl_loss()
loss = gdlloss(out,ctimgs1)

this is the gdl_loss

class gdl_loss(nn.Module):
def __init__(self, pNorm=2):
    super(gdl_loss, self).__init__()
    self.convX = nn.Conv2d(1, 1, kernel_size=(1, 2), stride=1, padding=(0, 1), bias=False)
    self.convY = nn.Conv2d(1, 1, kernel_size=(2, 1), stride=1, padding=(1, 0), bias=False)
    filterX = torch.FloatTensor([[[[-1, 1]]]])  # 1x2
    filterY = torch.FloatTensor([[[[1], [-1]]]])  # 2x1
    self.convX.weight = torch.nn.Parameter(filterX,requires_grad=False)
    self.convY.weight = torch.nn.Parameter(filterY,requires_grad=False)
    self.pNorm = pNorm

def forward(self, pct, ct):
    assert not ct.requires_grad
    assert pct.dim() == 4
    assert ct.dim() == 4
    assert pct.size() == ct.size(), "{0} vs {1} ".format(pct.size(), ct.size())
    pred_dx = torch.abs(self.convX(pct))
    pred_dy = torch.abs(self.convY(pct))
    gt_dx = torch.abs(self.convX(ct))
    gt_dy = torch.abs(self.convY(ct))
    grad_diff_x = torch.abs(gt_dx - pred_dx)
    grad_diff_y = torch.abs(gt_dy - pred_dy)
    mat_loss_x = grad_diff_x ** self.pNorm
    mat_loss_y = grad_diff_y ** self.pNorm  # Batch x Channel x width x height
    shape = ct.shape
    mean_loss = (torch.sum(mat_loss_x) + torch.sum(mat_loss_y)) / (shape[0] * shape[1] * shape[2] * shape[3])
    return mean_loss
2

You specified net.to(device), so the weights are in GPU memory , and the data type will be torch.cuda.FloatTensor.

However, you did not load your input to GPU, it’s still in CPU, which is type torch.FloatTensor.
So you should call something like:
mrimgs1 = mrimgs1.to(device)

3

Hi! thanks for your reply, I add it before the net,
it’s now:

mrimgs1 = mrimgs1.to(device)
ctimgs1 = ctimgs1. to(device)
out = net(mrimgs1)
gdlloss = gdl_loss()
gg = gdlloss(out,ctimgs1)

but the result is now:
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

4

That is because you did not add gdlloss to GPU.
gdlloss = gdlloss.to(device)

5

thanks sooooo much!!! it works now!!!
really appreciated it !!!

and now I know that the loss should also to Device :joy:

6

No problem.

To clarify, it’s because gdl_loss contains weights (torch.nn.Parameter), so you need to load them into GPU by calling to(device) before interacting them with inputs that are in GPU memory, just like what you did to net.

If you are using losses that does not define new weights, you don’t need to do so.

1 Like