Slightly unsure on the output of probabilities of semantic segmentation model.
>>> mask, decoded, probs = learn.predict(x) >>> probs.shape torch.Size([4, 360, 480])
My first guess was these are 3 channels of image and an alpha channel, but that doesn’t make sense when the prediction mask is only one channel.
My use case is I’d like to set any pixels with low probability to background class (ie: changing pixel label to 0)
Thanks for your help!