MEFaceNets: Muti-scale Efficient CNNs for Real-Time Face Recognition on Embedded Devices

The growing trend of applying face recognition technology on terminals and embedded devices highlights the critical need to strike a balance between recognition accuracy and real-time inference latency. In response to this challenge, we propose an efficient bottleneck named MEBottleneck, which utilizes convolution kernels of different sizes on two parallel branches to capture multi-scale features in the bottleneck, followed by a \(1 \times 1\) expansion layer to fuse multi-scale features, thereby improving the representation ability. Then, to balance the trade-off between accuracy and latency, we design a family of lightweight models with MEBottleneck, specifically tailored for face recognition and named MEFaceNets. Large kernels are used for depthwise convolutions in shallow layers, leading to notable improvements in accuracy. We evaluate the proposed models on several popular face recognition benchmarks. Our primary model achieves 99.80% face verification accuracy on LFW and exhibits excellent performance on the larger and more challenging benchmarks, including MegaFace Challenge 1, IJB-B and IJB-C. Furthermore, our proposed models have demonstrated impressive real-time performance on both the CPU and GPU of embedded devices.