Orthonormal Product Quantization Network for Scalable Face Image Retrieval

TL;DR

This paper introduces an orthonormal product quantization network that enhances face image retrieval by improving quantization informativeness, discriminability, and generalization, outperforming existing methods on multiple datasets.

Contribution

It proposes a novel end-to-end deep learning framework integrating orthonormal product quantization with a specialized loss and regularization to improve face retrieval accuracy and scalability.

Findings

Outperforms state-of-the-art deep hashing/quantization methods on face datasets.

Enhances model generalization to unseen identities.

Demonstrates broad applicability to general image retrieval tasks.

Abstract

Existing deep quantization methods provided an efficient solution for large-scale image retrieval. However, the significant intra-class variations like pose, illumination, and expressions in face images, still pose a challenge for face image retrieval. In light of this, face image retrieval requires sufficiently powerful learning metrics, which are absent in current deep quantization works. Moreover, to tackle the growing unseen identities in the query stage, face image retrieval drives more demands regarding model generalization and system scalability than general image retrieval tasks. This paper integrates product quantization with orthonormal constraints into an end-to-end deep learning framework to effectively retrieve face images. Specifically, a novel scheme that uses predefined orthonormal vectors as codewords is proposed to enhance the quantization informativeness and reduce codewords' redundancy. A tailored loss function maximizes discriminability among identities in each quantization subspace for both the quantized and original features. An entropy-based regularization term is imposed to reduce the quantization error. Experiments are conducted on four commonly-used face datasets under both seen and unseen identities retrieval settings. Our method outperforms all the compared deep hashing/quantization state-of-the-arts under both settings. Results validate the effectiveness of the proposed orthonormal codewords in improving models' standard retrieval performance and generalization ability. Combing with further experiments on two general image datasets, it demonstrates the broad superiority of our method for scalable image retrieval.