Web-Scale Training for Face Identification
This work addresses face recognition for security and identification applications, offering incremental improvements through novel insights into transfer learning and data sampling.
The study tackled face recognition by analyzing how deep convolutional networks transfer to large datasets, discovering that network bottlenecks act as regularizers, performance can saturate with more data, and representation norms affect discrimination; they improved LFW benchmark accuracy in both verification and identification protocols, showing a sizable leap over state-of-the-art commercial systems.
Scaling machine learning methods to very large datasets has attracted considerable attention in recent years, thanks to easy access to ubiquitous sensing and data from the web. We study face recognition and show that three distinct properties have surprising effects on the transferability of deep convolutional networks (CNN): (1) The bottleneck of the network serves as an important transfer learning regularizer, and (2) in contrast to the common wisdom, performance saturation may exist in CNN's (as the number of training samples grows); we propose a solution for alleviating this by replacing the naive random subsampling of the training set with a bootstrapping process. Moreover, (3) we find a link between the representation norm and the ability to discriminate in a target domain, which sheds lights on how such networks represent faces. Based on these discoveries, we are able to improve face recognition accuracy on the widely used LFW benchmark, both in the verification (1:1) and identification (1:N) protocols, and directly compare, for the first time, with the state of the art Commercially-Off-The-Shelf system and show a sizable leap in performance.