Allow users to help train the AI model

AI models requires reference examples; allow users to create custom models or help refine existing models by using their own reference material.

I think that is a very good idea.

A simple example would be: you have a photo shoot with one model in it. And you have shots that you want enhanced (let’s say Face AI, with the model behind a glass, where the person is not fully visible due to reflections etc.)
Now of course you would want the Face AI to reconstruct your model not an extrapolation of any person.
That could only be done, if one was able to train custom models for example for one photo set.

I suppose there would be at least two/three use cases

  • Train one/more faces for a specific photo set and tell face AI, which detected face (square) is which trained face
  • General model training, independent of the photo set. Every photographer / artist usually has a style, how photos should look in the end. Such a preference might be trained for general use
  • Improving pre-existing models, if users are not satisfied the way the currently perform. Let’s say you have a blur/tile issue and then you would have the base model (which every user has), and then you would also have an add-on file for each standard-model allowing users to add additional training. Such improvements could possibly even be brought back to new standard-models for all users, if shared with Topaz Labs.
1 Like