Query and database as sets/clusters of vectors

[AndrasZabo]

Hi All,
I'm new to vector similarity search, hopefully you can give some guidance to start my exploration in the good direction. Also apologies if I ask something obvious. My issue and question:

1. Let's say you extract hundreds of vectors describing one object in an image
2. Then you want to find the corresponding object (its vectors) in the database (lots of vectors describing different objects from different images (one object can be shown in several different images e.g. from different viewpoints)
3. Instead of searching for best matches for the hundreds of query vectors describing the current object in the image, does faiss provide an algorithm which represents the query "vector set" with a lower dimensional descriptor (e.g. only one vector, etc?) and finds its most similar "set/cluster of vectors" (the most similar object) in the database? Or are there some other approaches targeting/solving the same problem?

Thanks a lot if you can provide me some help! Also, do you think it's possible to implement this on edge devices? What may be the limiting factors? (computational power of the device, dimension of descriptor vectors, size of database?) Any general ideas, knowledge, tips are welcome. Thank you!