I plan to store 1-2 million of entities into a database. Each entity will be linked to 10-100 image embeddings. The goal is to use an external image that may or may not be part of the database and compare it to every stored embedding to find the most simmilar entity it could belong to.

I wish to ask, what kind of a database is most suitable for this task and why? How much slower would PostgreSQL with pgvector be compared to a more specialized DB such as Milvus or FAISS?