Computer VisionStatus: Production
BioAttend
Multi-modal biometric attendance — face, iris, and liveness in one pipeline
PythonFastAPIInsightFaceFAISSEfficientNet-B0OpenCV
The Problem
Standard face-match attendance systems are vulnerable to photo/video spoofing and lack a strong secondary verification factor.
The Solution & Architecture
Built a FastAPI-based multi-modal verification pipeline combining InsightFace Buffalo_L embeddings + FAISS for fast face search, iris verification as a second factor, and a custom EfficientNet-B0 liveness model trained on a self-collected video dataset to reject spoof attempts.
Pipeline Architecture Breakdown
1.Edge Camera Capture -> Frame Validation & Quality Check (OpenCV)
2.InsightFace Buffalo_L -> 512-D Feature Extraction & Normalization
3.FAISS Index FlatIP / HNSW -> Real-time Vector Similarity Retrieval
4.Iris Segmentation & ROI Extraction -> Polar coordinate unwrap & matching
5.EfficientNet-B0 Liveness Pipeline -> Binary classification with dynamic confidence thresholding
Core Highlights & Functionality
Multi-modal verification: High-accuracy facial recognition paired with secondary iris pattern matching
Anti-spoofing / Liveness: Custom EfficientNet-B0 classifier detecting print, screen replay, and mask attacks
Sub-second Vector Search: Indexed facial embeddings with FAISS across 10,000+ enrolled identities
High-throughput API: Asynchronous FastAPI architecture optimized for edge camera ingestion