Search code examples
deep-learningneural-networkgoogle-colaboratoryunet-neural-network

Unet predicts grey images (grey block)

I'm having problems with a unet-implementation and my data. The unet works with the testdata, but it doesn't work with my data. I only get a block of grey pixels as a prediction. I would like to use this unet for a university project. Does anybody see my fault?

here is the Repo I tried to copy: https://github.com/zhixuhao/unet

this is my Repo: https://github.com/Machuntox/unet

The membrane data is the test data and the JOSM data is for my project.

greetings, stefan

Solution

  • I found my error. I needed to preprocess my images differently. When I added the alpha channel with a value of 255, it worked. The last two steps did the trick.

    See commit list: https://github.com/Machuntox/unet/commits/master/prepareData.py