Using Method on real use case

[buenyamink]

Dear Author, i am interested in using few shot learning to collect images via few shot learning model for my thesis. I want to take one picture per class and these are used as support set. Afterwards it should move the next images, whish i am shooting, to the each classes folder.

How should i modify the Code to achive this iam a beginner

[cfifty]

Hey! I can suggest some code for the modeling part, but it would help if I understand the problem better. What are you trying to classify? What is the way (i.e. classification 5-ways would be classifying an unlabeled image as one of 5 different labels)?

[buenyamink]

Hello, thank you very much for the feedback.

The point is that it should simply be able to distinguish between different objects such as a screw, nut or a snickers etc. It is therefore not limited to specific objects. As I understand the paper, no meta learning on a related object is necessary. Am I right ? I will have one Image per class as support set (for example for 5 classes). In the end it should recognize a new image as one of these 5 classes. The number of classes can theoretically vary between 1 and 5.

[yaochx]

as I study the great work, I think you should do 3 things:

1. prepare your test data in MYOWNDATASET under ../caml_universal_eval_datasets directory
2. add new path config in utils.py's get_test_path() method
3. add cli arguments for src/evaluation/test.py, --eval_dataset MYOWNDATASET

3 you can find some tips here

@buenyamink