LEMoN: Label Error Detection using Multimodal Neighbors

TL;DR

LEMoN is a novel method that uses multimodal neighborhood information in pretrained models to detect label errors in image-caption datasets, improving data quality and downstream captioning performance.

Contribution

This paper introduces LEMoN, a new approach for label error detection in multimodal datasets, with theoretical justification and extensive empirical validation.

Findings

LEMoN outperforms existing baselines by over 3% in label error detection.

Filtering datasets with LEMoN improves downstream captioning by over 2 BLEU points.

Empirical validation across eight datasets confirms LEMoN's effectiveness.

Abstract

Large repositories of image-caption pairs are essential for the development of vision-language models. However, these datasets are often extracted from noisy data scraped from the web, and contain many mislabeled instances. In order to improve the reliability of downstream models, it is important to identify and filter images with incorrect captions. However, beyond filtering based on image-caption embedding similarity, no prior works have proposed other methods to filter noisy multimodal data, or concretely assessed the impact of noisy captioning data on downstream training. In this work, we propose, theoretically justify, and empirically validate LEMoN, a method to identify label errors in image-caption datasets. Our method leverages the multimodal neighborhood of image-caption pairs in the latent space of contrastively pretrained multimodal models to automatically identify label errors. Through empirical evaluations across eight datasets and twelve baselines, we find that LEMoN outperforms the baselines by over 3% in label error detection, and that training on datasets filtered using our method improves downstream captioning performance by more than 2 BLEU points over noisy training.