A synthetic dataset of 2D imagery, 3D point clouds, and 3D vehicle bounding box labels generated using the Grand Theft Auto 5 game engine.
GTA-3D Dataset is a synthetic dataset designed for 3D object localization research. It provides 2D imagery, 3D point clouds, and 3D vehicle bounding box labels generated using the Grand Theft Auto 5 game engine. The dataset solves the problem of scarce, accurately labeled 3D data for training computer vision models.
Researchers and developers working on 3D object detection, localization, and autonomous vehicle perception systems.
It offers a large-scale, high-resolution synthetic dataset with precise 3D labels, which is more cost-effective and scalable than collecting real-world data. The included Python utilities make it easy to integrate into machine learning pipelines.
A dataset of 2D imagery, 3D point cloud data, and 3D vehicle bounding box labels all generated using the Grand Theft Auto 5 game engine.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Offers 1680x1050 resolution images and depth maps, providing detailed visuals crucial for training accurate computer vision models.
Includes oriented 3D bounding box labels in both bird's-eye and image plane coordinates, enabling precise object localization for research.
Comes with `gta.py` helper classes that simplify data loading and processing, as demonstrated in the README with examples for RGB images and bounding boxes.
Leverages the GTA 5 game engine to generate large-scale data without the high costs and logistics of real-world data collection.
Synthetic data from a game engine lacks real-world sensor noise and environmental variations, which can hinder model generalization without adaptation.
At 55GB split into 11 ZIP files, downloading and managing the dataset requires substantial storage and can be tedious for quick experimentation.
Confined to GTA 5 urban environments and vehicle objects, missing diversity in weather, lighting, and non-vehicle elements common in real datasets.