How to fine tune a model from an MTCNN?

I’m making a system that can detect faces from artistic paintings. I use for this the face.evoLVe repo from

face.evoLVe uses an MTCNN for the facial detection and works brialliantly on real world images. However, I want to fine tune it using transfer learning to work on artistic painting such as the Mona Lisa.

I tried the tutorial on the PyTorch website,

However, I’m not very sure on where to include the model from face.evoLVe. Since it is an MTCNN, I assume I only have to fine tune the pnet.npy.

For the dataset on what it should fine tune I was thinking about having a dataset with two categories: faces and not-faces.