U-Net resulting in absolutely binary classification

Hello. I am rather new to Pytorch (used to work mostly with keras in previous jobs) and I am in need of some help. I am facing some issues when I implemented a segmentation model using the u-net from segmentation models

import segmentation_models_pytorch as smp

model_old = smp.Unet(
encoder_name=“resnet34”,
encoder_weights=None,
in_channels=3,
classes=8,
).to(device)

I train the model and it get quite ok results quite fast, but each pixel has always 0 or 1 at the final result. There is no degree of belief/pertinence of the pixel/area to the class. I really need the degree so I can apply some thresholds.

Is that expected from an U-NET, is is characteristic from this implementation or should I be doing something different to achieve that? Could any kind soul give me some pointers?