Learning to Hash Naturally Sorts

TL;DR

This paper introduces Naturally-Sorted Hashing (NSH), a novel deep hashing approach that aligns training with test metrics by differentiably sorting hash codes, improving unsupervised image retrieval performance.

Contribution

The paper proposes a differentiable sorting-based training method for deep hashing and introduces SortedNCE loss for unsupervised semantic relation mining.

Findings

NSH outperforms existing unsupervised hashing methods on benchmark datasets.

Differentiable sorting enables end-to-end training aligned with retrieval metrics.

SortedNCE effectively captures semantic relations during training.

Abstract

Learning to hash pictures a list-wise sorting problem. Its testing metrics, e.g., mean-average precision, count on a sorted candidate list ordered by pair-wise code similarity. However, scarcely does one train a deep hashing model with the sorted results end-to-end because of the non-differentiable nature of the sorting operation. This inconsistency in the objectives of training and test may lead to sub-optimal performance since the training loss often fails to reflect the actual retrieval metric. In this paper, we tackle this problem by introducing Naturally-Sorted Hashing (NSH). We sort the Hamming distances of samples' hash codes and accordingly gather their latent representations for self-supervised training. Thanks to the recent advances in differentiable sorting approximations, the hash head receives gradients from the sorter so that the hash encoder can be optimized along with the training procedure. Additionally, we describe a novel Sorted Noise-Contrastive Estimation (SortedNCE) loss that selectively picks positive and negative samples for contrastive learning, which allows NSH to mine data semantic relations during training in an unsupervised manner. Our extensive experiments show the proposed NSH model significantly outperforms the existing unsupervised hashing methods on three benchmarked datasets.