Once your prototypes are computed you need to average each class and normalize it, and add a the class names as a list under the key 'label_names'
Once you have done this, you can use demo/demo.py and changing category_space in the main function (you put the images on which you want to make inferences in the input dir, and you get the output in the output dir)
May I ask if there is any inference code provided and I would like to know the inference code for the feed shot.