Demo for image similarity project.

Features

1- Searching an image via text: Add positive and negative prompts to search for images. Avoid typing positive prompts such as 'not blue', instead, use the negative prompt.

2- Searching an image via image: Upload an image, and you will get similar images in the vector store.

Notes

• Qdrant is used as vector storage. Run Qdrant locally with Docker image to run the demo.
• CLIP, Titan and ALIGN models were used for the embeddings. These models can embed both text and image in semantic vector space, therefore we can search the similarities.

Running the Demo

• Run Qdrant on terminal: (colima start docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant)
• Run backend (app_clip_qdrant.py)
• Run frontend (frontend_clip.py)