# MTMMC: A Large-Scale Real-World Multi-Modal Camera Tracking Benchmark

**Authors:** Sanghyun Woo, Kwanyong Park, Inkyu Shin, Myungchul Kim, In So Kweon  
**arXiv:** [2403.20225](https://arxiv.org/abs/2403.20225)  
**Date:** 2024-03-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2403.20225v1/

## Key points

- Multi-target multi-camera tracking is crucial in visual surveillance, crowd behavior analysis, and anomaly detection.
- Existing datasets have limitations in modeling real-world dynamics and diverse camera configurations.
- The MTMMC dataset addresses this issue by offering a large-scale dataset captured by 16 multi-modal cameras in campus and factory environments.
- The dataset includes spatially aligned and temporally synchronized RGB and thermal cameras to enhance tracking accuracy.
- MTMMC benefits fields like person detection, re-identification, and multiple object tracking.
- Baselines and new learning setups are provided on this dataset for future studies.
- The datasets, models, and test server will be publicly available with users agreeing to the terms outlined in the Use Agreement.

## Summary

Multi-target multi-camera tracking is a critical task in various fields such as visual surveillance, crowd behavior analysis, and anomaly detection. The existing datasets for this task have limitations in modeling real-world dynamics and generalizing to diverse camera configurations due to their synthetic or artificially constructed nature. To address this issue, the MTMMC dataset has been introduced. It offers a large-scale dataset captured by 16 multi-modal cameras in campus and factory environments under different time, weather, and season conditions. This dataset provides a challenging test-bed for studying multi-camera tracking under diverse real-world complexities. It includes spatially aligned and temporally synchronized RGB and thermal cameras as an additional input modality, enhancing the accuracy of multi-camera tracking. MTMMC serves as a super-set of existing datasets and benefits independent fields like person detection, re-identification, and multiple object tracking. The authors provide baselines and new learning setups on this dataset while setting reference scores for future studies. The datasets, models, and test server will be made publicly available. Additionally, users must agree to the terms and conditions outlined in the Multi-Target Multi-Modal Camera Tracking Dataset Use Agreement before accessing the dataset. This refined summary highlights the significance of MTMMC in advancing research on multi-target multi-camera tracking in real-world scenarios.