How could I get the image feature from the n-th layer of the BLIP backbone?

[SwaggyPegg]

Thank you very much for your open source contribution, the performance of the model is amazing. If I want to obtain image features from the intermediate layers of the backbone, is there any related code that I can use directly?