Assuming you had a large corpus of 10 to 20 million images and wanted to do image similarity are there easy-to-work-with, stable indexes for vectors that provide make finding a set of “similar” vectors easy? I just need something that can…
- Take a single vector and compare it with all vectors in the index. Return a roughly ordered list.
- Allow for some sort of paginated results. I can get images that are most similar but then efficiently query for images that are less similar as I paginate to page 2, 3,4, etc.
I know Jeremy mentioned faiss a year or so ago in part 2. Has anyone worked with it? Are there alternatives out there that are better productized and easy to self-host in the cloud?