Hi,
I have a network (this is the model: PIPNet/pipnet at main · M-Nauta/PIPNet · GitHub) whose output is always positive, but I like to use this model for multi-label task where I have to use BCEwithlogitloss. But with this always-positive output, the built-in sigmoid function does not work. I like to know if someone has any recommendations.
My output is like this l = torch.tensor([[3.1847, 3.1812, 3.1875, 3.1981, 3.1937, 3.1888],
[3.4225, 3.4238, 3.4206, 3.4222, 3.4128, 3.4266],
[3.3188, 3.3171, 3.3155, 3.3196, 3.3170, 3.3163],
[3.2350, 3.2372, 3.2413, 3.2451, 3.2469, 3.2372],
[3.4386, 3.4308, 3.4364, 3.4415, 3.4254, 3.4318]])