Google researchers have published a paper on a facial recognition system that they say achieved nearly 100-percent accuracy on the facial recognition dataset Labelled Faces in the Wild (LFW).Called FaceNet, the researchers say the system uses a deep convolutional network to learn mapping from face images to a compact Euclidean space.The researchers say the approach is superior compared to other methods such as a CNN bottleneck layer, or those that require additional post-processing.The system achieved a classification accuracy of 98.87% on LFW when using a fixed center crop and a record breaking 99.63% when using an extra face alignment. LFW includes more than 13,000 pictures of faces from across the web.On YouTube Faces DB FaceNet also achieves 95.12%, and cuts the error rate in comparison to the best published result by 30% on both datasets.”The benefit of our approach is much greater representational efficiency: we achieve state-of-the-art face recognition performance using only 128-bytes per face”, writes the group.