Lighthouse: A User-Friendly Library for Reproducible Video Moment Retrieval and Highlight Detection

TL;DR

Lighthouse is a comprehensive, user-friendly library that standardizes and simplifies reproducible video moment retrieval and highlight detection research, enabling easier experimentation and application.

Contribution

It introduces a unified codebase with multiple models, features, and datasets, along with an inference API and web demo for accessible research and development.

Findings

Reproduces reported scores across various methods and datasets.

Facilitates easier setup and experimentation for researchers.

Provides a practical tool for advancing MR-HD research.

Abstract

We propose Lighthouse, a user-friendly library for reproducible video moment retrieval and highlight detection (MR-HD). Although researchers proposed various MR-HD approaches, the research community holds two main issues. The first is a lack of comprehensive and reproducible experiments across various methods, datasets, and video-text features. This is because no unified training and evaluation codebase covers multiple settings. The second is user-unfriendly design. Because previous works use different libraries, researchers set up individual environments. In addition, most works release only the training codes, requiring users to implement the whole inference process of MR-HD. Lighthouse addresses these issues by implementing a unified reproducible codebase that includes six models, three features, and five datasets. In addition, it provides an inference API and web demo to make these methods easily accessible for researchers and developers. Our experiments demonstrate that Lighthouse generally reproduces the reported scores in the reference papers. The code is available at https://github.com/line/lighthouse.