Unsupervised Memorability Modeling from Tip-of-the-Tongue Retrieval Queries

TL;DR

This paper introduces a large-scale unsupervised dataset with over 82,000 videos and recall data, enabling improved modeling of visual memorability and retrieval tasks without relying on expensive human annotations.

Contribution

It presents the first large-scale unsupervised dataset for visual memorability modeling, along with models that outperform existing methods in recall generation and tip-of-the-tongue retrieval.

Findings

Unsupervised dataset effectively models memorability signals.

Fine-tuned vision-language models outperform GPT-4o in description generation.

Contrastive training enables multimodal tip-of-the-tongue retrieval.

Abstract

Visual content memorability has intrigued the scientific community for decades, with applications ranging widely, from understanding nuanced aspects of human memory to enhancing content design. A significant challenge in progressing the field lies in the expensive process of collecting memorability annotations from humans. This limits the diversity and scalability of datasets for modeling visual content memorability. Most existing datasets are limited to collecting aggregate memorability scores for visual content, not capturing the nuanced memorability signals present in natural, open-ended recall descriptions. In this work, we introduce the first large-scale unsupervised dataset designed explicitly for modeling visual memorability signals, containing over 82,000 videos, accompanied by descriptive recall data. We leverage tip-of-the-tongue (ToT) retrieval queries from online platforms such as Reddit. We demonstrate that our unsupervised dataset provides rich signals for two memorability-related tasks: recall generation and ToT retrieval. Large vision-language models fine-tuned on our dataset outperform state-of-the-art models such as GPT-4o in generating open-ended memorability descriptions for visual content. We also employ a contrastive training strategy to create the first model capable of performing multimodal ToT retrieval. Our dataset and models present a novel direction, facilitating progress in visual content memorability research.