Synthetic Data Generation¶
This Section Contains¶
| Section | Description |
|---|---|
| Automatic Dump During Recording Playback | Collect simulation datasets via automatic Dump during recording playback |
| Dump Data Conversion | Convert raw Dump data into standard dataset formats |
| BEV Stitching Tool | Synthesize bird's-eye-view data using the BEV stitching tool |
| Sensor Dump Virtual Annotation Dataset | Format and field reference for the Sensor Dump annotation dataset |
How to Set Up Sensor Dump Data Collection¶
Related documentation: Sensor Dump Virtual Annotation Dataset
Dump
To meet customers' need to generate datasets through high-fidelity rendered scenes, we have added a Dump output mode for sensor data. When this output mode is selected, the data collected by the corresponding sensor and the ground-truth data for each frame (groundtruth) are stored in a specified directory during test case execution.
1. Introduction to Generating Dump Datasets from Standard Test Cases¶
In addition to test case execution, the SimOne simulation platform can provide labeled ground-truth data for perception algorithm testing and training.
- Camera Simulation: Provides annotated image datasets including semantic segmentation maps, depth maps, 2D/3D bounding boxes, supporting monocular, wide-angle, and other camera configurations.
- LiDAR Simulation: Provides raw point cloud data, annotated point cloud data, and bounding boxes of detected objects.
- Millimeter-Wave Radar Simulation: Provides target-level millimeter-wave radar detection data.
The following uses dumping camera perception data as an example. SimOne provides two Dump methods: manual Dump and automatic Dump. Dump results are provided as raw data; a parsed example is shown below:
