Pretrained model about ext_portrait_segmentation HOT 12 CLOSED

the pretrained models in dir /result/..

when I run Test_model.py to test the pretrained model，it occur errors.

File "E:/test/ext_portrait_segmentation-master/Test_model.py", line 158, in
lossVal, ElossVal, mIOU_val = val_edge(num_gpu, valLoader, model, criteria, Lovasz)
File "E:\test\ext_portrait_segmentation-master\etc\help_function.py", line 48, in val_edge
epoch_loss.append(loss.item())
UnboundLocalError: local variable 'loss' referenced before assignment

Do you know how to solve it?

from ext_portrait_segmentation.

the pretrained models in dir /result/..

from ext_portrait_segmentation.

yes, the pre-trained file is in result folder

from ext_portrait_segmentation.

😎 the file was too small, I didnot notice it

from ext_portrait_segmentation.

Thanks for great work. The pretrained model's results is not good as in the paper. Is it just a sample pretrained model for testing?

Thank you very much.

from ext_portrait_segmentation.

Could you say more detailed information?

from ext_portrait_segmentation.

@HYOJINPARK I built a test service based on your pretrained model, the results is not good:

from ext_portrait_segmentation.

@HYOJINPARK I tested with

from ext_portrait_segmentation.

Please check this code and result

Test_image.py.zip

from ext_portrait_segmentation.

the pretrained models in dir /result/..

when I run Test_model.py to test the pretrained model，it occur errors.

File "E:/test/ext_portrait_segmentation-master/Test_model.py", line 158, in
lossVal, ElossVal, mIOU_val = val_edge(num_gpu, valLoader, model, criteria, Lovasz)
File "E:\test\ext_portrait_segmentation-master\etc\help_function.py", line 48, in val_edge
epoch_loss.append(loss.item())
UnboundLocalError: local variable 'loss' referenced before assignment

Do you know how to solve it?

The following block of code in help_function.py was indented incorrectly:

with torch.no_grad():
            input_var, target_var, edge_target_var \
                = torch.autograd.Variable(input), torch.autograd.Variable(target), \
                    torch.autograd.Variable(edge_target)

            # run the mdoel
            output = model(input_var)

            # compute the loss
            loss = criterion(output, target_var)
            lossE = 0.5*criterion(output, edge_target_var)

It should be at the same level as

if num_gpu > 0:
            input = input.cuda()
            target = target.cuda()
            edge_target = edge_target.cuda()

Not inside that if statement.

from ext_portrait_segmentation.

Please check this code and result

Test_image.py.zip

Thanks for this code

from ext_portrait_segmentation.

Great. Too Small .pth.

from ext_portrait_segmentation.