Princeton Vision & Learning LabInfinigen is a procedural generator of 3D scenes optimized for computer vision research. This open-source tool generates diverse, high-quality 3D training data. With an active development community, Infinigen continuously expands its capabilities and coverage.
Key Features:
- Procedural Generation: Infinigen constructs 3D scenes entirely from randomized mathematical rules, encompassing shapes and materials at macro and micro levels, allowing for unlimited variations.
- Diverse Scene Generation: Infinigen covers a wide spectrum of natural objects and environments, including plants, animals, terrains, fire, clouds, rain, and snow, with future expansions planned for built environments and artificial objects.
- Real Geometry: By avoiding the use of bump/normal-maps and full-transparency techniques, Infinigen ensures all fine details of geometry are authentic, delivering accurate 3D ground truth for computer vision research.
- Automatic Annotations: Infinigen automatically generates high-quality annotations for various computer vision tasks, such as optical flow, 3D scene flow, depth, surface normals, and panoptic segmentation, while also offering customizable options to users.
