save problem

[HiYellowC]

how can I save the output as png or jpg?

[Honzys]

Use:

dot.format = format
dot.render(filename)

where

• dot is graph representation return using make_dot
• format is format from graphviz (see https://graphviz.readthedocs.io/en/stable/manual.html#formats)
• filename specifies path, where to save graph

[soulslicer]

I tried to do that and the image I get is nonetype:

            disp = model(imgL,imgR)

            dot = make_dot(disp) 
            print(dot)

            dot.render("test.png")

[kamalravi]

@soulslicer Try like this

x = torch.zeros(1, 3, 224, 224, dtype=torch.float, requires_grad=False)
out = model(x)
dot=make_dot(out)
print(dot)
dot.format = 'png'
dot.render("filename")

where

• x is your input
• model is your deep model (e.g. resnet)

You will get a "filename.png" image file. You can specify different format like 'svg' if you need.

[ucalyptus]

@kamalravi it gives
----> 7 dot.format("png")
8 dot.render("danet_model")

TypeError: 'str' object is not callable

[AruniRC]

I think a full example for saving might be helpful, especially for those not immediately familiar with graphviz.

dot = torchviz.make_dot(y.mean(), params=dict(model.named_parameters()))
dot.format = 'png'
dot.render('torchviz-sample')

And the image is saved under the current working directory as torchviz-sample.png.