Bad performance on low contrast images

[alvarofsan]

Hi!

I am doing some inference tests and I have realized that the model is not quite good when the object to be separated does not contrast too much with the background. Here is an example:

Original image:

Obtained mask:

As we can see below, when normalized to the range [-1, 1] with mean=0.5 and std=1 a much darker image is obtained. Could this affect the segmentation? Is there any way to improve this, maybe with postprocessing of the mask?

Normalized image:

I look forward to hearing from you, thank you very much for your work!