Open botoxparty opened 2 years ago
@botoxparty Hi, I think the links above give a great explanation of PTI. As for the question of multiple edits, the current code does not support it, but it should probably work.
As all editing directions are basically just numerical vectors, you could implement this by creating a new direction of age+gender fox example, by summing the age and gender directions.
You would need to create a new file in editings/w_directions that contains a weighted sum of age.npy and gender.npy
Hey there,
This is a really cool project. Managed to successfully apply this to some sample data that I have however now I am looking to tweak the results.
What is PTI and how does it affect the training?
Can it perform multiple edits, e.g. age + gender at once?