I created a meme search engine using siglip/CLIP and vector encoding images, learning a lot in the process. Vector embeddings have always fascinated me, and the magic of transforming text or images into numerical representations using techniques like Word2Vec is mind-blowing. Through my project, I delved into concepts like vector databases, OpenAI’s CLIP model, and different implementations like FAISS and ChromaDB. The process involved building a simple crawler to gather image embeddings, harnessing our Mac GPUs efficiently. The ability to search for similar images based on text descriptions is particularly intriguing and magical. The results are impressive, and this tech can revolutionize photo libraries and image searches. Check out my project on GitHub, and join me in exploring the endless possibilities of AI in imaging technology!
https://harper.blog/2024/04/12/i-accidentally-built-a-meme-search-engine/