Over 2 million traffic sign images were created by the OLIVES Lab at Georgia Tech to test algorithm robustness. Real-world images from BelgiumTS and synthetic images from Unreal Engine were processed with Adobe After Effects to simulate 12 challenging conditions like rain, snow, and blur. The dataset includes 14 sign types, such as speed limit and stop, with challenge severity levels ranging from 1 to 5.
Use Cases
- Testing model robustness based on simulated challenging conditions like rain, snow, and haze
- Benchmarking traffic sign classification algorithms based on 14 sign types including speed limit and stop
- Training models on synthetic data generated with Unreal Engine 4
- Analyzing performance degradation based on controlled challenge severity levels from 01 to 05
Strengths
- Dataset contains over 2 million images, providing a large-scale resource
- Includes 12 controlled synthetic challenging conditions such as decolorization, rain, and snow
- Covers 14 distinct traffic sign types, including speed limit, stop, and yield
- Combines both real-world images from BelgiumTS and synthetic images from Unreal Engine
Limitations
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is unknown, which may limit suitability assessment
- Last update date is unknown; freshness unverified
Provenance
- Source
- OLIVES Lab at Georgia Institute of Technology
- Collection Method
- Images cropped from the CURE-TSD dataset, with real-world images from BelgiumTS video sequences and synthetic images generated with Unreal Engine 4, then processed with Adobe After Effects.