Puqi7/MRVS_anomaly_long_video_dataset

Dataset Card for MRVS Anomaly Long Video Dataset for Public Safety Video Sensemaking

Dataset Summary

This dataset contains event-level annotations for a collection of long-form videos captured for public-safety-oriented video analysis research. Each annotated event includes a short event description, the associated video filename, the temporal range within the video, and the event duration in seconds.

The dataset is designed to support research on event understanding, temporal event localization, long-video review, and human-AI collaboration in multi-video sensemaking settings.

This dataset accompanies the paper:

Designing Multi-Robot Ground Video Sensemaking with Public Safety Professionals
Proceedings of the ACM Conference on Human Factors in Computing Systems (CHI 2026)

Supported Tasks and Applications

This dataset may be useful for research on:

• event-centric video understanding
• temporal event localization
• video review and retrieval
• public safety video analysis
• human-AI collaboration for video sensemaking
• multi-video investigation workflows

Dataset Structure

The repository contains two main components:

1. Raw videos stored directly in the repository
2. Event annotations stored in annotations/event_annotations.csv

Each event annotation is linked directly to a source video through its filename. No separate video index file is required.

Data Instances

Each row in annotations/event_annotations.csv corresponds to one annotated event.

Example:

Event	Video File	Start Time	End Time	Duration (sec)
Robots traffic	D1_1080_merged.mp4	03:37	04:15	38
Phone snatching	N2_1080p_merged.mp4	00:33	00:59	26

Data Fields

The event annotation file contains the following fields:

• Event: A short textual description of the annotated event.
• Video File: The filename of the source video associated with the event.
• Start Time: The start timestamp of the event within the source video.
• End Time: The end timestamp of the event within the source video.
• Duration (sec): The duration of the event in seconds.

Data Access and Usage

All source videos are stored directly in this repository.

Detailed event annotations are provided in:

• event_annotations.csv

A typical usage workflow is:

1. Load annotations/event_annotations.csv
2. Identify the event of interest
3. Locate the corresponding source video using the Video File field
4. Retrieve the relevant temporal segment using the provided Start Time and End Time

Annotation Process

The dataset was constructed through manual event annotation over a collection of long-form videos. Each event was annotated with:

• a short event description,
• the associated video filename,
• a start time,
• an end time,
• and the derived duration in seconds.

The annotations are intended to support event-level analysis rather than frame-level object detection or segmentation.

Source Data

The source data consists of long-form videos used for research on public safety video sensemaking. The videos are included directly in this repository and are referenced in the annotation file by filename.

Intended Uses

This dataset is intended for research and academic use, including:

• studying event-based video review workflows,
• evaluating temporal event retrieval methods,
• building interfaces for multi-video investigation,
• supporting research on public safety video sensemaking,
• exploring human-AI collaboration for event interpretation.

Out-of-Scope Uses

This dataset is not intended for:

• identity recognition or face recognition,
• biometric profiling,
• demographic inference,
• harmful surveillance practices,
• fully automated law-enforcement decision-making without human oversight.

Limitations

Several limitations should be considered:

• The dataset provides event-level annotations only.
• It does not include frame-level bounding boxes, segmentation masks, or dense object trajectories unless released separately.
• Event descriptions are concise and may not capture all contextual nuances.
• The dataset reflects the annotation scope and event definitions used in this project and may not generalize to all public safety scenarios.

Ethical Considerations

Because this dataset is related to video analysis in public-safety contexts, it should be used responsibly and only for legitimate research purposes.

Users should ensure that their use of the dataset complies with applicable privacy, data governance, and institutional review requirements. The dataset should not be used for harmful surveillance, discriminatory profiling, or operational decision-making without appropriate human oversight.

Repository Structure

• *.mp4: source videos included in the repository
• annotations/event_annotations.csv: detailed event annotations with video filename, time range, and duration
• README.md: dataset card and documentation

Licensing Information

Please specify the license that applies to:

1. the annotation files in this repository, and
2. the source videos included in this repository.

If the annotations and videos are governed by different usage terms, please describe them clearly here.

Citation

If you use this dataset, please cite:

@inproceedings{zhou2026designing,
  author = {Zhou, Puqi and Asgarov, Ali and Hussain, Aafiya and Park, Wonjoon and Paudyal, Amit and Shrestha, Sameep and Tang, Chia-Wei and Lighthiser, Michael and Hieb, Michael and Xiao, Xuesu and Thomas, Chris and Hong, Sungsoo Ray},
  title = {Designing Multi-Robot Ground Video Sensemaking with Public Safety Professionals},
  booktitle = {Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems},
  year = {2026},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  note = {To appear},
}