Combining tasks

I want to detect errors in images and I have the labels for the classification. If there is an error in the image I also have the location for that. So what I can do now is:

  • normal classification (good vs bad)
  • normal regression (location of error in an image, FlowField)

Now I want to combine these tasks. So what I propably have to do is:

  • adapt databunch
  • adapt network
  • adapt loss function

I guess I can manage point 2 and 3, but I have problems with point 1.
Can I use the MixedItemList class for that?

Thanks for helping out here.