Unity Perception

A toolkit for generating large-scale synthetic datasets in Unity to train and validate computer vision models.

Unity Perception streamlines the creation of high-fidelity synthetic data for machine learning workflows. By leveraging the High Definition Render Pipeline (HDRP), it produces pixel-perfect ground truth for tasks like 2D/3D object detection, semantic segmentation, and keypoint estimation. The package includes automated labeling tools (Labelers), randomized scene parameters (Domain Randomization), and the Perception Camera to export data in the standardized JSON-based Solo format. It eliminates the bottleneck of manual data collection by generating thousands of diverse, perfectly annotated images in minutes.

https://github.com/Unity-Technologies/com.unity.perception