How do extract features from a new architecture?

I am trying to extract features from a new multi-task learning architecture called MTI-net ( SimonVandenhende/Multi-Task-Learning-PyTorch: PyTorch implementation of multi-task learning architectures, incl. MTI-Net (ECCV2020). ( ).
Sadly, I only found tutorials for generally used architectures like Resnet …

Can some one point me into the right direction?