SeesawFaceNets: sparse and robust face verification model for mobile platform

Deep Convolutional Neural Network (DCNNs) come to be the most widely usedsolution for most computer vision related tasks, and one of the most importantapplication scenes is face verification. Due to its high-accuracy performance,deep face verification models of which the inference stage occurs on cloudplatform through internet plays the key role on most prectical scenes. However,two critical issues exist: First, individual privacy may not be well protectedsince they have to upload their personal photo and other private information tothe online cloud backend. Secondly, either training or inference stage istime-comsuming and the latency may affect customer experience, especially whenthe internet link speed is not so stable or in remote areas where mobilereception is not so good, but also in cities where building and otherconstruction may block mobile signals. Therefore, designing lightweightnetworks with low memory requirement and computational cost is one of the mostpractical solutions for face verification on mobile platform. In this paper, anovel mobile network named SeesawFaceNets, a simple but effective model, isproposed for productively deploying face recognition for mobile devices. Denseexperimental results have shown that our proposed model SeesawFaceNetsoutperforms the baseline MobilefaceNets, with only {\bf66\%}(146M VS 221MMAdds) computational cost, smaller batch size and less training steps, andSeesawFaceNets achieve comparable performance with other SOTA model e.g.mobiface with only {\bf54.2\%}(1.3M VS 2.4M) parameters and {\bf31.6\%}(146M VS462M MAdds) computational cost, It is also eventually competitive againstlarge-scale deep-networks face recognition on all 5 listed public validationdatasets, with {\bf6.5\%}(4.2M VS 65M) parameters and {\bf4.35\%}(526M VS 12GMAdds) computational cost.