search

bowang-lab / U-Mamba

U-Mamba: Enhancing Long-range Dependency for Biomedical Image Segmentation
https://arxiv.org/abs/2401.04722
Apache License 2.0
674 stars 60 forks source link

Got negative prediction value #63

Open Younger330 opened 5 hours ago

Younger330 commented 5 hours ago

Hello, I used your model for training in our Trainer, but the prediction is negative Here is my code. Am I wrong? Or would you have any suggestions? `images=images.unsqueeze(4) out=umamba.forward(images)

    pred=out[0].squeeze(2)
    pred=F.log_softmax(pred,dim=1)
    #the predict is negative

`

Younger330 commented 5 hours ago

The problem is got by model.apply(InitWeights_He(1e-2)), but I dont know how it influence, maybe only used for radiological images?

Younger330 commented 42 minutes ago

The problem is got by model.apply(InitWeights_He(1e-2)), but I dont know how it influence, maybe only used for radiological images?

Still have this problem after some iterations.