othmaneirl/Maritime_Visual_Tracking_Dataset_MVTD

---
license: cc0-1.0
size_categories:
- 100K<n<1M
---
# MVTD: Maritime Visual Tracking Dataset

## Overview

**MVTD (Maritime Visual Tracking Dataset)** is a large-scale benchmark dataset designed specifically for **single-object visual tracking (VOT) in maritime environments**.
It addresses challenges unique to maritime scenes: such as water reflections, low-contrast objects, dynamic backgrounds, scale variation, and severe illumination changes—which are not adequately covered by generic tracking datasets.

The dataset contains **182 annotated video sequences** with approximately **150,000 frames**, spanning **four maritime object categories**:
- Boat
- Ship
- Sailboat
- Unmanned Surface Vehicle (USV)

MVTD is suitable for **training, fine-tuning, and benchmarking** visual object tracking algorithms under realistic maritime conditions.

---

## Dataset Statistics

- **Total sequences:** 182
- **Total annotated frames:** 150,058
- **Frame rate:** 30 FPS and 60 FPS
- **Resolution range:**
- Min: 1024 × 1024
- Max: 1920 × 1440
- **Average sequence length:** ~824 frames
- **Sequence length range:** 82 – 4747 frames
- **Object categories:** 4

---

## Dataset Structure

The dataset follows the **GOT-10k single-object tracking format**, enabling easy integration with existing tracking pipelines.

---

MVTD/
├── train/
│ ├── video1/
│ │ ├── frame0001.jpg
│ │ ├── frame0002.jpg
│ │ ├── ...
│ │ ├── groundtruth.txt
│ │ ├── absence.label
│ │ ├── cut_by_image.label
│ │ └── cover.label
│ ├── video2/
│ │ ├── frame0001.jpg
│ │ ├── frame0002.jpg
│ │ ├── ...
│ │ ├── groundtruth.txt
│ │ ├── absence.label
│ │ ├── cut_by_image.label
│ │ └── cover.label
│ └── ...
└── test/
├── video1/
│ ├── frame0001.jpg
│ ├── frame0002.jpg
│ ├── ...
│ └── groundtruth.txt
├── video2/
│ ├── frame0001.jpg
│ ├── frame0002.jpg
│ ├── ...
│ └── groundtruth.txt
└── ...

---

## Tracking Attributes

Each video sequence is categorized using **nine tracking attributes**:

1. Occlusion
2. Illumination Change
3. Scale Variation
4. Motion Blur
5. Variation in Appearance
6. Partial Visibility
7. Low Resolution
8. Background Clutter
9. Low-Contrast Objects

These attributes represent both **maritime-specific** and **generic VOT challenges**.

---

## Data Collection

The dataset was collected using **two complementary camera setups**:

- **Onshore static camera**
- Large scale variations
- Perspective distortions
- Occlusions from vessels and structures

- **Offshore dynamic camera mounted on a USV**
- Strong illumination changes and glare
- Motion blur and vibrations
- Rapid viewpoint changes

This setup covers diverse maritime scenarios including:
- Coastal surveillance
- Harbor monitoring
- Open-sea vessel tracking

---
## Evaluation Protocols

MVTD supports two evaluation settings.
For detailed implementation, evaluation scripts, and baseline tracker configurations, please visit the official GitHub repository:

🔗 **https://github.com/AhsanBaidar/MVTD**

### Protocol I – Pretrained Evaluation
- Trackers pretrained on generic object tracking datasets
- Evaluated directly on the MVTD test split
- Measures generalization performance in maritime environments

### Protocol II – Fine-Tuning Evaluation
- Trackers fine-tuned using the MVTD training split
- Evaluated on the MVTD test split
- Measures domain adaptation effectiveness for maritime tracking

## Baseline Results

The dataset has been benchmarked using **14 state-of-the-art visual trackers**, including Siamese, Transformer-based, and autoregressive models.
Results show **significant performance degradation** when using generic pretrained trackers and **substantial gains after fine-tuning**, highlighting the importance of maritime-specific data.

---

## Intended Use

MVTD is suitable for:
- Single-object visual tracking
- Domain adaptation and transfer learning
- Maritime robotics and autonomous navigation
- Benchmarking tracking algorithms under maritime conditions

---

## Citation

If you use this dataset, please cite:

```bibtex
@article{bakht2025mvtd,
title={MVTD: A Benchmark Dataset for Maritime Visual Object Tracking},
author={Bakht, Ahsan Baidar and Din, Muhayy Ud and Javed, Sajid and Hussain, Irfan},
journal={arXiv preprint arXiv:2506.02866},
year={2025}
}