I have implemented a U-Net, similar to this implementation, but for a different dataset, this one, to segment roads.

It works fine using the test folder images, but, for example, when I pick a print from bing maps and try to infer with the trained model, this is returned:

enter image description here

Why this is happening?

I already tried to change the thresholding values, normalization, etc.


  • 1
    $\begingroup$ One thing to check - image could be with value range in [0, 255] and [0, 1]. Make sure this range is identical for train and test images) $\endgroup$ Mar 17, 2021 at 15:31
  • $\begingroup$ When you say in the title "with images different from the dataset", you mean "test dataset", right? It may be a good idea to show us an example of an image from the test dataset too (even though you already provide a link to the dataset). $\endgroup$
    – nbro
    Mar 18, 2021 at 10:30


You must log in to answer this question.