CPU

[luantunez]

Hello, I was trying to use your model on skin segmentation in a CPU, but I am not being able to predict over an image. Is that possible? Would you help me with the code modifications, please?
Thank you, Lucia

[WillBrennan]

There's three places where .cuda() or device='cuda' is used. These need to be a global variable somewhere; that defines what device to use.

device = 'cuda' if torch.cuda.is_available() else 'cpu'

I need to make this change to the two other pytorch projects as well. I'll look at doing it in the next week.

[luantunez]

Thank you for your response. I was able to make the predictions with your modifications. I am not getting the same results though, but a much smaller threshold is needed and not even then I get a good segmentation.
Do you know why should that be?

[luantunez]

BTW, which are these three places? I am only seeing two and getting the runtime error though:

device = torch.device("cpu")

being used in:
model = load_model(models[model_type], torch.load(model_ckpt, map_location=device))
image = image.to(device).unsqueeze(0)

[WillBrennan]

There was also a call to model.cuda().eval() which needed to be changed. You can see it in the PR #17