How to get image featrue using my pre-trained model?

Which feature/layer would you like to get?
You could use forward hooks as shown in this example or modify your forward function to return a specific activation.