HNC: Leveraging Hard Negative Captions towards Models with Fine-Grained Visual-Linguistic Comprehension Capabilities

TL;DR

This paper introduces HNC, a dataset of hard negative captions and a benchmark test set to improve fine-grained visual-linguistic understanding in image-text matching models.

Contribution

It presents a novel dataset of hard negative captions and a challenging benchmark for enhancing and evaluating fine-grained cross-modal comprehension.

Findings

Training with HNC improves zero-shot mismatch detection.

HNC models perform robustly with noisy visual inputs.

HNC provides better initialization for fine-tuning.

Abstract

Image-Text-Matching (ITM) is one of the defacto methods of learning generalized representations from a large corpus in Vision and Language (VL). However, due to the weak association between the web-collected image-text pairs, models fail to show a fine-grained understanding of the combined semantics of these modalities. To address this issue we propose Hard Negative Captions (HNC): an automatically created dataset containing foiled hard negative captions for ITM training towards achieving fine-grained cross-modal comprehension in VL. Additionally, we provide a challenging manually-created test set for benchmarking models on a fine-grained cross-modal mismatch task with varying levels of compositional complexity. Our results show the effectiveness of training on HNC by improving the models' zero-shot capabilities in detecting mismatches on diagnostic tasks and performing robustly under noisy visual input scenarios. Also, we demonstrate that HNC models yield a comparable or better initialization for fine-tuning