search

ybendou / easy

This repository is the official implementation Ensemble Augmented-Shot Y-shaped Learning: State-Of-The-Art Few-Shot Classification with Simple Ingredients.
MIT License
112 stars 18 forks source link

Saving and deploying the Few-Shot model #17

Closed rlleshi closed 2 years ago

rlleshi commented 2 years ago

Are we also saving the few-shot head here?

I am concerned with deploying the model in production. Looks like we can't just do a normal prediction on the loaded model but need to take care of the logic of few_shot_eval.eval_few_shot?

ybendou commented 2 years ago

Hello,

Sorry for the delayed answer,

In our method we don't use a few-shot head since we use a Nearest Class Mean approach. Indeed, to deploy in production, you would need first to pass the images through your model, then use the following function .

We have a version which is more production-like for a demo in the following repository, you can check it to get an inspiration.

Hope this help, let me know if it doesn't.

Best,

rlleshi commented 2 years ago

Thanks a lot for getting back! That answers the question.