Tensorboardx: How to use the function add_graph(model, input_to_model, verbose=False)? (It seems that demo_graph.py is not executable)

[MrStrangerYang]

At the beginning, I tried to run demo_graph.py, but the result was always

Then I tried to add a main function code:

if __name__ == '__main__':
    exp_name = 'model_' + time.strftime("%y-%m-%d_%H-%M-%S", time.localtime()) + 'resnet'
    w = SummaryWriter('./run' + '/' + exp_name)

    dummy_input = Variable(torch.rand(1, 3, 224, 224))
    model = torchvision.models.resnet18()
    w.add_graph(model, dummy_input)

And started up the tensorboard using the command:

tensorboard --logdir ./run --port 6611

However the result showed in my browser was

So, I wonder to know the correct way to use the function add_graph to draw the network structure diagram like this:

[lanpa]

This screenshot is very old.(for pytorch 0.2)

please check #46 for more information. I suggest to use add_graph() along with pytorch 0.4 for best view. (And help me verifying the result lol)

[JiamingSuen]

I figured out that tensorboardX must be installed from source in order to use add_graph(), with pytorch > 0.4. Maybe this should be put into readme as well.

[lanpa]

@JiamingSuen Thanks, I think it should be mentioned. And FYI, the pip package was updated yesterday.

[Anubhav7]

@JiamingSuen Thanks for pointing out that this was due to version mismatch. Works after upgrading pytorch to 0.4