Iterating through a set of images and getting a score for a keyword (solved)

[ghost]

There was a website that did exactly that, but it didnt display a score, it displayed the three images that matched the keyword (and connected CLIP embedding) the most. Wish I could find it (and its code)

Lets say I have a set of images and I want to ask clip how much of a "dog" is in there, and it should give me a number. How would I go about realising this? Of course, a CLIP model needs to be loaded and accessed in python. I know all the basics, about embeddings and the sampling process, but I am not a very experienced python coder and putting this puzzle piece together is really really difficult for me.

I got an aesthetic scorer going but that was only possible because I was able to use code parts from someone else.

Anyone have an idea about a project that does this, that I can modify? Or where do I even begin to code this from scratch?

edit: Solved. For anyone interested: this here does the trick and even more:

https://github.com/shonenkov/CLIP-ODS

https://openai.com/blog/clip/

(sorry for using an openai link here :D)