Can this detection library be used for training?

I have pip this package, And the effect is very good.

But I want to know if this can be used for training?

For example, I want to predict the landmarks of the generated image and the input image, and calculate the landmark loss to train the network.

The issue asks if the face-alignment library can be used for training, specifically to compute landmark loss. Start by examining the library's API and source code to see if it exposes the necessary components (like landmark predictions) for integration into a training loop. Look for examples or documentation on model outputs and check if gradients can flow through the library's operations. Determine what modifications or wrapper code would be needed to use it within a PyTorch training script.

Written by the indexing model from the issue text.