Semantic segmentation masks exported are all black

I am creating a U-nets model. When I want to export my annotation dataset as a polygon, all masks are black.

I recently saw that you have released a new video related to SegFormer. It’s good, but with all black masks I can guarantee that the model will not learn anything!


Hi they’re not actually all black. The pixel values are Eg 0, 1, 2, 3, etc which all look close to black on your screen. The model is perfectly able to differentiate these values from each other though.

Check out the visualization code in the notebook that remaps to human interpretable colors.

Okay, I didn’t understand it that way. Yes, everything is clear now.
Thanks a lot!

1 Like