Improving Face Verification Accuracy Using Hyperplane Similarity
Dr. Mike Jones, Senior Principal Research Scientist, Mitsubishi Electric Research Labs
Wednesday, December 13, 1:00-2:00pm; MCS 148 (111 Cummington Mall, Boston)
Abstract: The standard framework for using a convolutional neural network (CNN) for face verification is to compare the feature vectors taken from the penultimate network layer of a CNN trained to classify the identity of an input face using a softmax loss over identities. Feature vectors are typically compared using the simple L2 distance. We demonstrate that the L2 distance is not the best distance to use in this scenario, and propose the hyperplane similarity as a more appropriate similarity function that is derived from the softmax loss function used to train the network. We demonstrate that hyperplane similarity improves verification results especially for low false acceptance rates which are usually the most important operating regimes for real applications. We also propose a fast algorithm for finding the separating hyperplanes needed to compute hyperplane similarity.
Bio: Mike Jones earned his Ph.D. in 1997 from the Artificial Intelligence Lab at MIT under the supervision of Tomaso Poggio for his work on 2D morphable models. After graduating he joined Digital Equipment Corporation’s Cambridge Research Lab and then moved to Mitsubishi Electric Research Labs (MERL) in 2001 where is currently a Senior Principal Research Scientist. He is best known for his work with Paul Viola on face detection using AdaBoost which won the Longuet-Higgins Prize at CVPR in 2011. He also won the Marr Prize at ICCV in 2004 for work on pedestrian detection.