|
|
--- |
|
|
task_categories: |
|
|
- token-classification |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- automobile |
|
|
- sensor |
|
|
pretty_name: vot-rgbt |
|
|
size_categories: |
|
|
- 100M<n<1B |
|
|
--- |
|
|
|
|
|
# 🚁 VOT-RGBT 2019 Challenge Dataset |
|
|
|
|
|
**Visual Object Tracking for RGB and Thermal Imagery** |
|
|
|
|
|
--- |
|
|
|
|
|
## Dataset Summary |
|
|
|
|
|
The **VOT-RGBT 2019** dataset is part of the [Visual Object Tracking](http://www.votchallenge.net/) (VOT) initiative — a series of challenges that provide the computer vision community with standardized benchmarks for evaluating **short-term** and **long-term visual object trackers**. This particular edition focuses on **RGB-T (Visible + Thermal)** imagery, encouraging robust object tracking in challenging multimodal environments. |
|
|
|
|
|
--- |
|
|
|
|
|
## Supported Tasks and Leaderboards |
|
|
|
|
|
- **📦 Visual Object Tracking (Short-Term & Long-Term)** |
|
|
- Track object location across RGB and thermal frames |
|
|
- Evaluate robustness to occlusion, illumination changes, and environmental conditions |
|
|
|
|
|
--- |
|
|
|
|
|
## Dataset Structure |
|
|
|
|
|
- **Sequences**: 60+ sequences with aligned RGB and thermal imagery |
|
|
- **Annotations**: Per-frame ground truth bounding boxes in both modalities |
|
|
- **Modalities**: RGB, Thermal (LWIR) |
|
|
- **Frame Rate**: 20-30 FPS (varies by sequence) |
|
|
- **Resolution**: Varies by sensor and sequence (mostly HD and VGA) |
|
|
|
|
|
--- |
|
|
|
|
|
## Usage |
|
|
|
|
|
To use the dataset: |
|
|
|
|
|
```python |
|
|
from datasets import load_dataset |
|
|
# This assumes dataset is hosted on Hugging Face datasets hub |
|
|
dataset = load_dataset("langutang/vot-rgbt2019") |
|
|
``` |
|
|
|
|
|
Alternatively, you can download it from the [official VOT site](http://www.votchallenge.net/vot2019/). |
|
|
|
|
|
--- |
|
|
|
|
|
## Evaluation Protocol |
|
|
|
|
|
VOT-RGBT 2019 uses the standard VOT evaluation protocol: |
|
|
|
|
|
- **Accuracy (A)**: Overlap between predicted and ground truth bounding boxes |
|
|
- **Robustness (R)**: Number of tracking failures |
|
|
- **Expected Average Overlap (EAO)**: Combines A and R for a unified score |
|
|
|
|
|
Both short-term and long-term tracker performance can be evaluated using the provided toolkit. |
|
|
|
|
|
--- |
|
|
|
|
|
## Citation |
|
|
|
|
|
If you use this dataset in your research, please cite: |
|
|
|
|
|
``` |
|
|
@inproceedings{votrgbt2019, |
|
|
title={RGB-Thermal Object Tracking: Benchmark and Baselines}, |
|
|
author={Liang, Jianan and Hu, Jiakai and Zhang, Yu and others}, |
|
|
booktitle={ECCV Workshops}, |
|
|
year={2019} |
|
|
} |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## License |
|
|
|
|
|
Please refer to the license and usage terms outlined on the official VOT website. |
|
|
|
|
|
--- |
|
|
|
|
|
## Acknowledgements |
|
|
|
|
|
Thanks to the VOT committee and contributing authors for their continued efforts in pushing forward the field of visual object tracking. |
|
|
|
|
|
--- |
|
|
|
|
|
## Tags |
|
|
|
|
|
`computer-vision`, `object-tracking`, `rgbt`, `thermal-imaging`, `vot`, `multimodal`, `benchmark` |
|
|
|
|
|
|
