ImageSearch
Search a folder of pictures by describing what you're looking for. A small vector search built on CLIP.
- When
- 2026
- Built with
- Python, CLIP, ChromaDB
ImageSearch finds pictures by meaning. You type "a red bicycle" and get the pictures that look like one, even though nothing in their filenames says so.
It uses CLIP, a model that puts pictures and sentences into the same space, so a sentence lands near the pictures it describes. Every image is turned into a vector once and stored in a vector database. A search turns your words into a vector the same way and returns the nearest images.
It is deliberately simple, two files, and is set up to try on the Caltech-256 image set.