FE-DETRAC Dataset (Fisheye8K)
Welcome to the FE-DETRAC benchmark dataset.
The dataset is provided to support academic research on fisheye-camera object detection, multi-object tracking, intelligent transportation systems, and related computer vision applications.
Citation Requirement
If you use any part of the FE-DETRAC dataset in a publication, report, thesis, presentation, website, software repository, demonstration, or other research output, you must cite the FE-DETRAC paper listed below.
This citation requirement applies to the use of:
- original videos or video clips;
- individual video frames, images, or photographs;
- cropped, resized, enhanced, or otherwise processed images;
- screenshots or visual examples derived from the dataset;
- object annotations, labels, bounding boxes, and metadata;
- training, validation, or test subsets;
- converted annotation formats;
- statistics, benchmark results, or dataset descriptions;
- modified, transformed, or derivative versions of any dataset content.
Use of only a portion of the dataset does not remove the citation requirement. Any redistribution or public display of FE-DETRAC data, images, photographs, annotations, or derivative materials must retain appropriate attribution to the original dataset and its associated publication.
📖 Citation
Plain Text — IEEE Style
C.-I. Huang, W.-Y. Chen, J.-W. Hsieh, P.-Y. Chen, M.-C. Chang, M. Gochoo, C.-M. Tsai, H.-W. Huang, V. S. Tseng, and J.-N. Hwang, “FE-DETRAC: An Infrastructure Fisheye Video Benchmark with Distortion-Aware Hybrid Data Association for Multi-Object Tracking,” IEEE Internet of Things Journal, 2026, doi: 10.1109/JIOT.2026.3711126.
BibTeX
bibtex
@article{huang2026fedetrac,
title = {FE-DETRAC: An Infrastructure Fisheye Video Benchmark with Distortion-Aware Hybrid Data Association for Multi-Object Tracking},
author = {Huang, Chung-I and Chen, Wei-Yu and Hsieh, Jun-Wei and Chen, Ping-Yang and Chang, Ming-Ching and Gochoo, Munkhjargal and Tsai, Chun-Ming and Huang, Hsiang-Wei and Tseng, Vincent S. and Hwang, Jenq-Neng},
journal = {IEEE Internet of Things Journal},
year = {2026},
doi = {10.1109/JIOT.2026.3711126},
note = {Early Access}
}
📊 Dataset Features
The FE-DETRAC dataset was constructed to support research under the substantial scale variation and hemispherical distortion produced by infrastructure-mounted fisheye cameras. Particular attention was given to small and distant road users, including scooters and pedestrians, whose appearance may vary considerably across image regions.
The dataset provides:
- Large-scale annotations: 20,000 continuous video frames with more than 470,000 annotated bounding boxes.
- Five road-user categories: Pedestrian, Bike, Car, Bus, and Truck.
- Real-world traffic scenes: Data collected using 22 fixed fisheye IoT cameras installed at intersections in Hsinchu City, Taiwan.
- High-resolution imagery: Video resolutions of 1920 × 1080 and 1920 × 1920 pixels.
- Multiple annotation formats: Pascal VOC, COCO, MOT, and YOLO.
- Camera-disjoint evaluation: Training and test sets are separated by camera to reduce scene-specific leakage and support a more rigorous evaluation of model generalization.
⚖️ Copyright and License
The FE-DETRAC dataset is distributed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).
By accessing, downloading, using, modifying, displaying, or redistributing any portion of the dataset, you agree to the following conditions.
1. Attribution and Citation
You must provide appropriate attribution to the FE-DETRAC authors and cite the associated paper in any research output that uses the dataset or any of its contents.
This requirement covers data, annotations, videos, video frames, photographs, cropped images, screenshots, visual examples, converted files, processed data, and derivative materials.
The required citation is:
C.-I. Huang, W.-Y. Chen, J.-W. Hsieh, P.-Y. Chen, M.-C. Chang, M. Gochoo, C.-M. Tsai, H.-W. Huang, V. S. Tseng, and J.-N. Hwang, “FE-DETRAC: An Infrastructure Fisheye Video Benchmark with Distortion-Aware Hybrid Data Association for Multi-Object Tracking,” IEEE Internet of Things Journal, 2026, doi: 10.1109/JIOT.2026.3711126.
When dataset images or photographs are reproduced in a paper, presentation, website, repository, poster, or other public material, the figure caption or accompanying text should identify them as originating from the FE-DETRAC dataset and include the corresponding paper citation.
2. Non-Commercial Use
The dataset may not be used for commercial purposes without prior written permission from the copyright holders.
For commercial use or licensing inquiries, please contact the authors.
3. ShareAlike
If you remix, transform, annotate, or build upon the dataset, you may distribute the resulting materials only under the same CC BY-NC-SA 4.0 license.
4. Preservation of Attribution
Copyright notices, license information, dataset identifiers, and citation instructions included with the original release must not be removed or obscured when the dataset or its derivative materials are redistributed.
Disclaimer
The FE-DETRAC dataset is provided “as is,” without warranties of any kind, express or implied. The authors and affiliated institutions shall not be held responsible for any loss, damage, claim, or consequence arising from access to or use of the dataset.
Users are responsible for ensuring that their use of the dataset complies with applicable laws, institutional requirements, ethical standards, and the terms of the CC BY-NC-SA 4.0 license.
