Project Overview
Offline AI desktop app for face tagging and semantic image search, processing 2.5–3 images/sec on a 4-core CPU with YuNet, SFace, CLIP, and FAISS.
An Electron interface talks to a local FastAPI Python service running a 3-stage ONNX Runtime pipeline (YuNet, SFace 128D, CLIP 512D) indexed in FAISS and SQLite.
Processes 2.5–3 images/sec on a 4-core CPU (1,000 photos indexed in ~6.5 minutes) fully offline on your machine.
Engineering decision
An Electron interface talks to a local FastAPI Python service running a 3-stage ONNX Runtime pipeline (YuNet, SFace 128D, CLIP 512D) indexed in FAISS and SQLite.
Outcome
Processes 2.5–3 images/sec on a 4-core CPU (1,000 photos indexed in ~6.5 minutes) fully offline on your machine.
Key Features
- Face detection and grouping on your own machine
- Search photos by description with CLIP embeddings
- FAISS index reused across later searches
- Electron interface backed by a local Python service