VIDI: A Video Dataset of Incidents

TL;DR

This paper introduces VIDI, a comprehensive video dataset of 43 incident categories, and demonstrates that using video data significantly improves incident classification accuracy over still images.

Contribution

The paper presents VIDI, a new diverse video dataset for incident detection, and benchmarks state-of-the-art models showing the benefits of video data for classification.

Findings

Video data improves incident classification accuracy from 67.37% to 76.56%.

Recent models like Vision Transformer and TimeSformer outperform previous approaches.

VIDI dataset will be publicly available for further research.

Abstract

Automatic detection of natural disasters and incidents has become more important as a tool for fast response. There have been many studies to detect incidents using still images and text. However, the number of approaches that exploit temporal information is rather limited. One of the main reasons for this is that a diverse video dataset with various incident types does not exist. To address this need, in this paper we present a video dataset, Video Dataset of Incidents, VIDI, that contains 4,534 video clips corresponding to 43 incident categories. Each incident class has around 100 videos with a duration of ten seconds on average. To increase diversity, the videos have been searched in several languages. To assess the performance of the recent state-of-the-art approaches, Vision Transformer and TimeSformer, as well as to explore the contribution of video-based information for incident classification, we performed benchmark experiments on the VIDI and Incidents Dataset. We have shown that the recent methods improve the incident classification accuracy. We have found that employing video data is very beneficial for the task. By using the video data, the top-1 accuracy is increased to 76.56% from 67.37%, which was obtained using a single frame. VIDI will be made publicly available. Additional materials can be found at the following link: https://github.com/vididataset/VIDI.