How ahmednasser5/kemet-vision works
This repository provides an Egyptian-artifact image recognition experience: a user submits a photo, the application checks whether it resembles the known artifact domain, classifies accepted images, and presents the prediction and confidence details. The router also exposes image-to-query and entity-tracking behavior for a conversational/RAG workflow. The README describes a three-model CLIP verification ensemble and NLP-generated descriptions, but the available source excerpts do not show those implementations or their wiring, so those capabilities are represented as documented, unverified branches. The README also describes a mobile app, but no mobile source is present in this tree.
Main entry points: multimodal_router.py EfficientNet classifier · GUI.py Prediction results
Components 14
Recognition Interface
- Streamlit interface
GUI.py - Prediction results
GUI.py - Mobile application
Image Recognition
- Image loading
multimodal_router.py - Camera detection
multimodal_router.py - OOD similarity filter
multimodal_router.py - EfficientNet classifier
multimodal_router.py - CLIP verifiers
- Ensemble vote
Conversation Context
- RAG query generation
multimodal_router.py - Entity extraction
multimodal_router.py - Entity session store
multimodal_router.py - Artifact description
Other components
- Artifact visitor
Connections 18
- Artifact visitor → Streamlit interface: submits photo
README.md - Streamlit interface → Image loading: loads image
- Image loading → Camera detection: checks source
multimodal_router.py - Camera detection → OOD similarity filter: selects threshold
multimodal_router.py - OOD similarity filter → EfficientNet classifier: accepts image
multimodal_router.py - EfficientNet classifier → CLIP verifiers: low confidence verifies
README.md - CLIP verifiers → Ensemble vote: casts votes
README.md - EfficientNet classifier → Ensemble vote: fallback prediction
README.md - EfficientNet classifier → Prediction results: returns prediction
- Ensemble vote → Prediction results: returns consensus
README.md - Prediction results → Artifact visitor: shows result
- Image loading → RAG query generation: generates query
multimodal_router.py - RAG query generation → Entity extraction: tracks entity
multimodal_router.py - Entity extraction → Entity session store: updates entity
multimodal_router.py - Artifact visitor → Artifact description: asks about artifact
README.md - EfficientNet classifier → Artifact description: provides class
README.md - Artifact description → Artifact visitor: returns description
README.md - Artifact visitor → Mobile application: uses app
README.md