Synset Boulevard: 259,200 Synthetic Front-View Vehicle Images

Fraunhofer IOSB released this collection of 259,200 synthetic images in 2024 for surveillance-based vehicle make and model recognition (VMMR). The data was generated using path tracing and physically-based modeling to depict 162 vehicle models from 43 different makes in front-view orientations.

Use Cases

• Training image classification models to identify 43 different vehicle makes
• Fine-grained recognition of 162 specific vehicle models in surveillance-style settings
• Benchmarking image segmentation performance on synthetic vehicle data

Strengths

• 259,200 high-fidelity synthetic images
• Covers 162 distinct vehicle models across 43 makes
• Physically-based path tracing ensures realistic lighting and material properties

Limitations

• Restricted to front-view perspectives only
• Synthetic-to-real domain gap may affect performance on actual low-resolution surveillance footage

Provenance

Source

Fraunhofer IOSB (Sielemann et al., 2024)

Collection Method

Synthetic generation via path tracing and physically-based modeling

Time Range

2024

Freshness

The dataset was last updated on March 13, 2026.