AggNet: Learning to Aggregate Faces for Group Membership Verification

TL;DR

This paper introduces AggNet, a deep learning architecture that jointly learns face descriptors and aggregation for group membership verification, enabling flexible and improved verification performance on large-scale datasets.

Contribution

It proposes a novel end-to-end deep architecture that adapts to new groups and memberships, outperforming existing quantization-based methods.

Findings

Higher verification accuracy on large-scale wild-face datasets

Effective handling of new group memberships and individual variations

Outperforms baseline aggregation methods

Abstract

In some face recognition applications, we are interested to verify whether an individual is a member of a group, without revealing their identity. Some existing methods, propose a mechanism for quantizing precomputed face descriptors into discrete embeddings and aggregating them into one group representation. However, this mechanism is only optimized for a given closed set of individuals and needs to learn the group representations from scratch every time the groups are changed. In this paper, we propose a deep architecture that jointly learns face descriptors and the aggregation mechanism for better end-to-end performances. The system can be applied to new groups with individuals never seen before and the scheme easily manages new memberships or membership endings. We show through experiments on multiple large-scale wild-face datasets, that the proposed method leads to higher verification performance compared to other baselines.