Compute similarity score against any given generation

[caviri]

Hello,

I was wondering if you updated the repository with the compute similarity score against any given generation feature? If not, I would like to help on this if possible.

Thanks a lot for your work,
Carlos

[learn2phoenix]

@caviri Do you want to compute top-N similar embeddings from the database given your embedding?

[caviri]

Hi @learn2phoenix, yes I would like to obtain the distance of a target image providing a set of reference images.

Also, I would like to develop a Dockerfile a provide and API endpoint for inference. Do you accept contributions?

[learn2phoenix]

I'll assume you want to get the closest reference images from the database (the case where you have the reference images yourself is trivial, let me know if you want to discuss that too).

To get the list of n reference images and their corresponding distance to the target image, you could use

CSD/search/faiss_search.py

Line 21 in 3a9df32

return self.index.search(embeddings, n)

to get the embeddings of reference images. Computing L2 distance to your target embedding is trivial from here.

We would love to accept a PR for a dockerfile. Could you clarify what you mean by API Endpoint?