Semantic Image Clustering with CLIP and K-Means

Objective

Built a full-stack prototype that automatically groups uploaded images by visual-semantic similarity and reports clustering quality.

Stack

Delivery highlights

• Developed an image-clustering workflow that accepts multiple uploaded images, generates semantic embeddings with CLIP (ViT-B/32), and groups them into a user-selected number of clusters using K-Means. Calculated a silhouette score to indicate clustering quality, exposed the pipeline through a FastAPI endpoint, and built a React interface that renders each result as a grouped image gallery. The current prototype demonstrates automatic visual organization but does not yet persist or export finished albums.