How mridulpant2010/face_recognition works
This repository combines a face-recognition RAG pipeline with standalone neural-network and margin-loss implementations. The principal workflow is to discover and split identity-labeled images, embed gallery faces, persist a searchable index and probe set, then retrieve identity matches for queries; the CLI also exposes evaluation. The README is empty, and evaluation, configuration, and some model files were not sampled, so their internal behavior and wiring are not inferred.
Main entry points: run_pipeline.py Pipeline CLI · ingest.py Ingestion · arcface_embedder.py ArcFace embedder · faiss_vector_store.py FAISS store
Components 18
RAG pipeline
- Pipeline CLI
src/face_recognition/rag/run_pipeline.py - Ingestion
src/face_recognition/rag/pipeline/ingest.py - Face retriever
src/face_recognition/rag/pipeline/retrieve.py - Evaluation
src/face_recognition/rag/pipeline/evaluate.py - Pipeline configuration
src/face_recognition/rag/core/config.py
Embedding
- Embedder interface
src/face_recognition/rag/interfaces/base_embedder.py - ArcFace embedder
src/face_recognition/rag/implementations/arcface_embedder.py
Model components
- IResNet architecture
src/face_recognition/iresnet.py - Vision Transformer
src/face_recognition/vit.py - Margin losses
src/face_recognition/losses.py - Image-text model
src/face_recognition/imagetextmodel.py
Index and data
- Store interface
src/face_recognition/rag/interfaces/base_vector_store.py - FAISS store
src/face_recognition/rag/implementations/faiss_vector_store.py - Identity image dataset
src/dataset - Saved index and metadata
- Saved probe set
Other components
- CLI user
- InsightFace runtime
Connections 22
- CLI user → Pipeline CLI: runs
- Pipeline CLI → Ingestion: runs ingestion
- Pipeline CLI → Evaluation: runs evaluation
- Pipeline CLI → Face retriever: queries via
- Pipeline CLI → ArcFace embedder: creates
- Pipeline CLI → FAISS store: creates and loads
- Ingestion → Identity image dataset: discovers images
- Ingestion → ArcFace embedder: embeds gallery
- Ingestion → FAISS store: adds and saves
- Ingestion → Saved probe set: writes probe paths
- ArcFace embedder → InsightFace runtime: runs inference
- Face retriever → ArcFace embedder: embeds query
- Face retriever → FAISS store: searches
- FAISS store → Saved index and metadata: reads and writes
- Ingestion → Pipeline configuration: reads settings
- Face retriever → Pipeline configuration: reads defaults
- ArcFace embedder → Embedder interface: implements
- FAISS store → Store interface: implements
- Pipeline CLI → IResNet architecture: not evidenced
- Pipeline CLI → Vision Transformer: not evidenced
- Pipeline CLI → Margin losses: not evidenced
- Pipeline CLI → Image-text model: not evidenced