---
license: mit
tags:
- depth
- autodrive
- odometry
- disparity
pretty_name: The ETS2 Dataset
size_categories:
- 10K<n<100K
---
# ETS2 Dataset: Synthetic Data from Euro Truck Simulator 2 for Monocular Depth Estimation

[![PyTorch](https://img.shields.io/badge/PyTorch-ee4c2c?logo=pytorch&logoColor=white)](#)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

## 📝 Dataset Description

The ETS2 Dataset is a large-scale synthetic dataset generated using Euro Truck Simulator 2 (ETS2), a popular driving simulation game. It captures high-resolution driving behavior, sensor data, and environmental context across a variety of roads, traffic conditions, and driving styles.

The dataset was created for research in autonomous driving, driver behavior modeling, and realistic vehicle control systems in simulation-based environments. By leveraging the visual and physics fidelity of ETS2, this dataset bridges the gap between synthetic and real-world driving data — while offering full control, reproducibility, and scalability.


- **46,780 RGB images** (1440×816 resolution)
- **Dense depth maps** (2-byte short float format)
- **Vehicle telemetry data** (position, velocity, acceleration, etc.)
- **Session metadata** (road type, weather, time of day, traffic level)

The dataset is organized into **18 driving sessions**, each with unique environmental conditions to ensure diversity and robustness for training deep learning models.

## 📊 Statistics

- **Total frames**: 46,780  
- **Stopped frames**: 1,659 (3.55%)  
- **Weather**: 88.78% sunny, 11.22% rainy  
- **Road types**: Urban (48.93%), Highway (23.01%), Road (16.42%), Industrial (11.63%)

## 🧪 Benchmarking

We validated ETS2 by training two state-of-the-art models:
- **Monodepth2** (self-supervised)
- **DenseDepth** (supervised)

Both models trained on ETS2 were evaluated on the **KITTI** dataset. Results show that synthetic training can generalize to real-world data, with finer-grained predictions observed in ETS2-trained models.

## 🔧 Tools and Code

- 📦 Dataset: [Hugging Face Dataset Page](https://huggingface.co/datasets/dmariaa70/ets2-dataset)
- 🛠️ Data Capture Tool: [ets2-data-capture](https://github.com/dmariaa/ets2-data-capture)
- 🧪 Experiments: [ets2-trainer](https://github.com/dmariaa/ets2-trainer)

## 📄 License

This project is licensed under the **MIT License**.

You are free to:

- ✅ Use the dataset for commercial and non-commercial purposes  
- ✅ Modify, distribute, and build upon the dataset  
- ✅ Incorporate it into your own models, research, or applications  

As long as you include the following notice in your work:

> **Copyright © 2023 David María Arribas <david.maria@urjc.es>**  
> This software and dataset are provided "as is", without warranty of any kind, express or implied.

The full license text is available [here](https://opensource.org/licenses/MIT).

## 📖 Citation

If you use this dataset in your research, please cite the following paper:

```bibtex
@article{Mara-Arribas2023,
   abstract = {In this work, we present a new dataset for monocular depth estimation created by extracting images, dense depth maps, and odometer data from a realistic video game simulation, Euro Truck Simulator 2 TM. The dataset is used to train state-of-the-art depth estimation models in both supervised and unsupervised ways, which are evaluated against real-world sequences. Our results demonstrate that models trained exclusively with synthetic data achieve satisfactory performance in the real domain. The quantitative evaluation brings light to possible causes of domain gap in monocular depth estimation. Specifically, we discuss the effects of coarse-grained ground-truth depth maps in contrast to the fine-grained depth estimation. The dataset and code for data extraction and experiments are released open-source.},
   author = {David María-Arribas and Alfredo Cuesta-Infante and Juan J. Pantrigo},
   doi = {10.1007/978-3-031-36616-1_30},
   isbn = {978-3-031-36616-1},
   issn = {16113349},
   journal = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
   keywords = {depth estimation,domain gap,simulation,synthetic dataset,video games},
   month = {6},
   pages = {375-386},
   publisher = {Springer Nature},
   title = {The ETS2 Dataset, Synthetic Data from Video Games for Monocular Depth Estimation},
   volume = {14062 LNCS},
   url = {https://hdl.handle.net/10115/90437},
   year = {2023}
}