Facial Recognition / CLIP Smart Search models comparison / benchmarks

[Rundervinkje]

Hello there,
I've been searching for an hour now (general internet, this page, discord) and thought i'd just ask for help now.
I was wondering if there is an overview/comparison of the available models and their intended use-case, accuracy, resource intensity and other possible metrics that would be usefull in deciding the right model for my case. (raspi 5, headless with docker, Dutch)

Thank you in advance for your thoughts / help

[mertalev]

The small and large variants are benchmarked here if it helps. My recommendation is to just stick to the default unless you can’t run it.

[Rundervinkje]

@mertalev thank you for the quick response. Do you know if there is a similar page for the smart detection models? Then i'll rename this discussion post for future answer-lookers?
I've been looking through the https://github.com/huggingface/transformers/tree/main but i honestly don't know what terms to search for.

[mertalev]

There's a chart here (multilingual version) that goes into detail. It isn't that easy to read, though.

The guide here has a chart I made based on this data and profiling. The models there are the ones I think are candidates to be the best (for English) within their size category.

[Rundervinkje]

You're a godsend! I'll read trough those. Happy holidays!
Edit: haha I see a familiar name on these posts!