Robust face representation is imperative to highly accurate face recognition.In this work, we propose an open source face recognition method with deeprepresentation named as VIPLFaceNet, which is a 10-layer deep convolutionalneural network with 7 convolutional layers and 3 fully-connected layers.Compared with the well-known AlexNet, our VIPLFaceNet takes only 20% trainingtime and 60% testing time, but achieves 40\% drop in error rate on thereal-world face recognition benchmark LFW. Our VIPLFaceNet achieves 98.60% meanaccuracy on LFW using one single network. An open-source C++ SDK based onVIPLFaceNet is released under BSD license. The SDK takes about 150ms to processone face image in a single thread on an i7 desktop CPU. VIPLFaceNet provides astate-of-the-art start point for both academic and industrial face recognitionapplications.
展开▼