Multimodal Memorability: Modeling Effects of Semantics and Decay on Video Memorability

TL;DR

This paper introduces a new model for predicting how video memories decay over time, integrating visual and semantic data, and demonstrates its effectiveness on a novel dataset and benchmarks.

Contribution

The paper presents Memento10k, a dynamic video memorability dataset, and a novel decay model that estimates memory retention over arbitrary delays using multimodal information.

Findings

Model outperforms previous approaches by 12% on average.

Introduces the first quantitative decay estimation for videos.

Combines visual and semantic data for improved predictions.

Abstract

A key capability of an intelligent system is deciding when events from past experience must be remembered and when they can be forgotten. Towards this goal, we develop a predictive model of human visual event memory and how those memories decay over time. We introduce Memento10k, a new, dynamic video memorability dataset containing human annotations at different viewing delays. Based on our findings we propose a new mathematical formulation of memorability decay, resulting in a model that is able to produce the first quantitative estimation of how a video decays in memory over time. In contrast with previous work, our model can predict the probability that a video will be remembered at an arbitrary delay. Importantly, our approach combines visual and semantic information (in the form of textual captions) to fully represent the meaning of events. Our experiments on two video memorability benchmarks, including Memento10k, show that our model significantly improves upon the best prior approach (by 12% on average).