edgeface
EdgeFace is an efficient face recognition model designed for edge devices, featuring a compact architecture suitable for resource-constrained environments. Published in IEEE T-BIOM (2024), it won the compact track of the IJCB 2023 Efficient Face Recognition Competition. The repository provides inference code and pretrained models for generating face embeddings from aligned face images. EdgeFace is available in multiple variants including base, small (s), extra small (xs), and ultra small (xxs), ranging from 18.23M to 1.24M parameters with computational costs from 1398 to 94 MFLOPs. Quantized versions with reduced storage are also available. Performance benchmarks show strong accuracy across standard datasets including LFW (99.57-99.83%), CALFW (94.83-96.07%), CPLFW (90.27-93.75%), CFP-FP (93.63-97.01%), and AgeDB30 (94.92-97.60%), demonstrating effective trade-offs between model size and recognition accuracy. Models can be loaded via PyTorch hub for easy integration. The implementation supports face alignment