In the world of autonomous vehicles and intelligent speed assistance (ISA) systems, detecting and classifying a wide variety of road signs is crucial. Ensuring perception models can handle even the most challenging scenarios requires extensive, diverse, and realistic training data.
With the European Union making ISA mandatory for all new vehicles by 2024, the demand for enhanced speed sign detection and classification models has never been higher. Parallel Domain is here to help, providing high-quality synthetic data that significantly improves road sign detection and classification performance.
An array of road signs brought to life: Parallel Domain's synthetic data showcases diverse sign types with 2D bounding box annotations
Our synthetic data has proven to make a real difference in perception performance, particularly in meeting the growing demand for ISA compliance. When using our data, customers have experienced an average of +15% mAP (mean average precision) improvement for classifying EU and other speed signs. In cases where real data is limited, we’ve even seen performance boosts of up to 50%.
A demonstrative example of a customer's performance gains achieved with Parallel Domain's synthetic sign data
Our platform supports several hundred signs used for roads throughout the United States, Japan, and European regions. We’re constantly expanding our library to cover more scenario types, geographical features, and regions. In addition to standard annotations, we can provide metadata about sign color, text, shape, and country-specific sign codes.
Synthetic ISA sign examples: accurately representing german road signs for enhanced perception models
For classification training, we can provide densely placed signs on maps, allowing you to use cropped sign data without context. For detection training, we offer more realistically distributed signs, enabling the use of full frames to detect sign locations and feed the information downstream to your classification model.
Suburban sign symphony: diverse synthetic road sign examples seamlessly integrated into a neighborhood setting
To ensure our synthetic data closely mirrors real-world situations, we offer customizable wear and tear options for certain types of signs placed throughout our maps. These customizations encompass rust, grime, damage, and varied reflectivity.
Parallel Domain can quickly generate datasets with high distributions of signage, addressing the need for diverse training data. Moreover, our platform can generate data for countries where data collection is challenging, such as Cyprus or Slovakia.
Synthetic highway scene at night: accurately representing lighting and sign reflectivity
Parallel Domain’s synthetic data platform offers a powerful solution for improving road sign detection and classification performance in autonomous vehicles and ISA systems. With our high-quality synthetic data, diverse sign support, and tailored data for different training needs, we’re spearheading the drive towards more intelligent and secure AI perception.
Eager to supercharge your sign detection and classification? Contact us for a sample of our high-impact synthetic data and see the difference firsthand. Connect with the Parallel Domain team today!