What's the different between the add_graph function's arguments: input_to_model

[Anonymous-so]

In add_graph function,there is a arguments needed: input_to model.When I go and check the source code, the args comments are follows:

input_to_model (torch.Tensor or list of torch.Tensor): A variable or a tuple of variables to be fed.

But when i try 3 different types of input_tensor(input_tensor, input_tensor1) and pass them to the add_graph function.The tensorboardX show 2 different GRAPHS.: writer = SummaryWriter(log_dir='runs/test-Res50/test-tensor0') net = torchvision.models.resnet50(pretrained=True) input_tensor = torch.zeros(10,3,224,224) input_tensor1 = (torch.zeros(10,3,224,224),) writer.add_graph(net, input_tensor) #or writer.add_graph(net, input_tensor1)

Does anybody know the difference between two input_tensor, and why the GRAPHS they produce are different?

the input_tensor's GRAPH picture: http://i2.tiimg.com/698066/5755785396df8792.png the input_tensor1's GRAPH picture:http://i2.tiimg.com/698066/2c4af29e936f0d2c.png

[lanpa]

Hi, this might be a bug that has been fixed. I tested with torch 1.2 and master branch of tensorboardX, the two inputs have the same rendering result.