Milestone Systems introduced Hafnia synthetic data capabilities and a forthcoming Training-as-a-Service offering at NVIDIA GTC in San Jose, enabling developers to train vision AI models for rare and previously unseen scenarios alongside real-world conditions.
The Hafnia platform connects data curation, model training, and deployment in a single environment. Developers can use it to reduce dataset bias and build computer vision models for smart city applications. The platform now integrates NVIDIA’s Physical AI Data Factory Blueprint as its reference architecture.
Hafnia Synthetic Data Addresses Real-World Gaps
AI systems typically learn from historical data, which often underrepresents rare weather conditions, unusual traffic patterns, and region-specific vehicle types. Hafnia addresses this limitation by integrating synthetic data into its curated real-world video library through NVIDIA Cosmos Transfer.
The synthetic augmentation layer gives developers access to data covering rare or dangerous situations, balanced object classes, regional and environmental variations, and systematically reduced dataset biases. Milestone stated that synthetic data enhances rather than replaces the platform’s real-world foundation, preserving authenticity, compliance, and annotation quality.
“Together with NVIDIA, we are taking Hafnia to the next level by combining trusted real-world data with synthetic augmentation. This enables developers to train AI models that are not only accurate in known situations, but also resilient in the unexpected.”
Edward Mauser, Director of Hafnia, Milestone Systems
Training-as-a-Service Previewed at GTC
Milestone previewed its upcoming Training-as-a-Service at the conference. The offering provides developers with a direct path from Hafnia’s data library to training infrastructure, removing the need to manage fragmented data pipelines. Developers can customize datasets and fine-tune models for specific use cases within their own training environments.
Because all data within Hafnia’s library is compliantly sourced and fully traceable, developers can train models with confidence regarding regulatory compliance. Milestone stated the service can accelerate the development of analytics solutions by up to 30 times by reducing data management complexity.
VLM-as-a-Service Performance Results
In partnership with NVIDIA, Hafnia also offers VLM-as-a-Service, a suite of Visual Language Models built on NVIDIA Cosmos Reason models and optimized for smart city environments. At GTC, Milestone stated the availability of a new EU-optimized VLM for traffic analysis, already deployed with selected European cities.
Performance evaluations show measurable improvements over base models across three metrics.
- +19.4% improvement in flow and direction correctness
- +8.9% improvement in visual feature detection
- +4.4% improvement in alert verification accuracy
Milestone stated the VLM-as-a-Service offering can accelerate the final stages of the model development lifecycle by up to 70 times. Additional models covering more smart city scenarios are in development.
Multi-Cloud Infrastructure and Data Sovereignty
Hafnia operates on a multi-cloud strategy using AWS, Nebius, and other providers to support the complete vision model lifecycle. Milestone launched its Synthetic Data Generation pipeline, powered by Cosmos Transfer and Cosmos Evaluator, on Nebius at GTC.
The multi-cloud approach supports data sovereignty requirements, giving customers full control over where their sensitive data is stored and processed. This architecture spans initial data sourcing, fine-tuning, full-scale training, and deployment across the complete artificial intelligence model lifecycle.
GTC Presence and Upcoming Presentation
Visitors to NVIDIA GTC can view live demonstrations of Hafnia’s expanded data library, preview Training-as-a-Service, and explore VLM capabilities at Booth #2036. Edward Mauser, Director and Product Lead at Milestone Systems, presented “A New Frontier for Vision AI with Expert Reasoning Agents” on March 18, from 2:00 to 2:40 PM.
Milestone Systems, founded in 1998 and headquartered in Copenhagen, employs more than 1,500 people worldwide. The company has been an independent entity within the Canon Group since 2014. Its product suite includes XProtect video management software, BriefCam analytics, and Arcules cloud VSaaS, alongside the Hafnia data compliance and anonymization platform.