Here you can download the official Toolkit for the SCOPE dataset. The Toolkit provides functionality for an efficient parallel download, visualization and evaluation of Collective Perception Algorithms.
Use the package manager pip to install the Toolkit.
pip install scope-toolkit
Once installed, you can start using the Toolkit to enhance your workflow with the SCOPE dataset.
The Toolkit allows efficient parallel downloading and extraction into the correct structure of the SCOPE dataset.
Simply run
cd ~/dataset-toolkit
python dataset-toolkit/TODO_ADD_PATH_TO/download_dataset.py
and follow the on-screen instructions.
The visualization module allows for an easy exploration of the dataset.
To visualize a scenario run
cd ~/dataset-toolkit
python dataset-toolkit/TODO_ADD_PATH_TO/visualize_sequence.py
and follow the on-screen instructions. This will create a GIF of the selected scenario while following the selected car in the given scenario.
Currently there are a few different perspectives implemented. Those are
graph TD
A{Viewpoint} -->|third-person/roof Camera+LiDAR| B(roof-option)
A --> |bird's eye view 45° only LiDAR| C[GIF]
B -->|front view| D(camera alignment)
B -->|front to back| E(camera alignment)
D --> |next to| F[GIF]
D --> |above| G[GIF]
E --> |next to| H[GIF]
E --> |above| I[GIF]
Evaluation module for easy and fair comparison of different methods using state-of-the-art metrics TODO
Our data structure optimises data management and analysis with unique features:
It separates environmental conditions, ensuring independent assessment of weather conditions. The hierarchical arrangement includes sensor data and transformation matrices per vehicle per scenario, enabling systematic management. Coordinate transformation is simplified with pre-existing matrices for each vehicle and sensor. In addition, support for common file formats such as .bin, .txt and .png ensures efficient data loading.
This section outlines the structure and contents of the ground truth data stored in a CSV (Comma-Separated Values) format. The table provides an overview of the fields present in the CSV file.
Each column is a separate field, such as the following
m_type << "," << m_id << "," << m_realworld_pos.x() << "," << m_realworld_pos.y() << "," << m_realworld_pos.z() << "," << m_realworld_dim.width <<...
The different types are
Type ID | Description |
---|---|
0 | BACKGROUND |
1 | CAR |
2 | VAN |
3 | TRUCK |
4 | PEDESTRIAN |
5 | PERSON_SITTING |
6 | CYCLIST |
7 | TRAM |
8 | MISC |
9 | MOTORBIKE |
If you use our SCOPE dataset or the toolkit, please use the following citation