Normalized Contrastive Learning for Text-Video Retrieval

TL;DR

This paper identifies a normalization issue in cross-modal contrastive learning for text-video retrieval and proposes NCL with Sinkhorn-Knopp to improve fairness and performance, achieving state-of-the-art results.

Contribution

The paper introduces Normalized Contrastive Learning (NCL) using Sinkhorn-Knopp to properly normalize retrieval probabilities in multimodal retrieval tasks.

Findings

NCL improves retrieval performance across multiple datasets.

NCL achieves new state-of-the-art metrics without architecture changes.

Normalization correction significantly benefits cross-modal retrieval accuracy.

Abstract

Cross-modal contrastive learning has led the recent advances in multimodal retrieval with its simplicity and effectiveness. In this work, however, we reveal that cross-modal contrastive learning suffers from incorrect normalization of the sum retrieval probabilities of each text or video instance. Specifically, we show that many test instances are either over- or under-represented during retrieval, significantly hurting the retrieval performance. To address this problem, we propose Normalized Contrastive Learning (NCL) which utilizes the Sinkhorn-Knopp algorithm to compute the instance-wise biases that properly normalize the sum retrieval probabilities of each instance so that every text and video instance is fairly represented during cross-modal retrieval. Empirical study shows that NCL brings consistent and significant gains in text-video retrieval on different model architectures, with new state-of-the-art multimodal retrieval metrics on the ActivityNet, MSVD, and MSR-VTT datasets without any architecture engineering.