AI and Data Security in Leica Cyclone 3DR

In reality capture, AI is having a profound impact, especially on the classification process. AI can now automatically identify and segment features within point clouds, which was previously a time-consuming, manual office task. However, with the increasing adoption of AI in reality capture software, concerns around data security are growing because AI tools typically require cloud connectivity.

In this Q&A, Yannick Stenger, Reality Capture Software Product Manager, explains how Leica Cyclone 3DR handles AI differently – keeping customer data offline and secure.

Should customers be worried about data safety when using AI tools in Leica Cyclone 3DR?

No, and here’s why. Many of our customers – especially in enterprise – are understandably cautious about AI, often requesting certifications and forms before allowing AI into their workflows.

But with Cyclone 3DR, those concerns don’t apply. The software, and the AI within it, run completely offline.

Unlike many AI solutions in the market, which rely on constant connectivity to cloud platforms like AWS or Azure, the AI tools in Cyclone 3DR are designed to function without internet connectivity, entirely on local hardware.

Don’t worry, this doesn’t reduce performance – the AI models are optimised to run seamlessly while offline. We recently ran a benchmarking test on a Dell laptop equipped with NVIDIA RTX GPUs powered by battery, with no online connection. The results were seamless.

Our offline-first approach to Cyclone 3DR is a key advantage for customers prioritising data security.

Is any customer data ever used to train Hexagon’s AI models?

No live or customer-specific data is used to train our AI models unless the customer has explicitly agreed to it. The AI in Cyclone 3DR is trained using internal datasets collected by Hexagon from select, older customer projects – but only where a formal agreement or NDA is in place. 

For example, some customers with HxDR accounts choose to share their data with us for development purposes.

However, any data processed locally and offline in Cyclone 3DR remains private and is never sent back to Hexagon or used for training.

What if a customer wants an AI point cloud classification (PCC) model trained on their own data?

That’s a very common request, especially from customers working in niche sectors. For example, one recent rail customer wanted an AI model trained on the specific layout and signs of their national rail system.

We can support customers like this looking to train a custom AI model, but they would need to share their data with us to train it – which wouldn’t be offline.

Still, our general models achieve around 80–85% accuracy, which is sufficient for most use cases. We have AI models for a wide range of real-world environments, such as rail, construction sites, roads, and buildings. We have also trained models for specific applications, like scan-to-pipe for MEP workflows and progress monitoring for AEC.

One exciting offline AI feature which comes with the latest release of Cyclone 3DR is Smart Object Segmentation. It allows users to isolate objects in a point cloud by simply tracing over the shape with their mouse. This reduces the need for time-consuming manual selection.

Following an initial AI PCC step, which provides an automated first pass, Smart Object Segmentation then offers user-defined refinement.

Compared to the AI PCC models, this new feature is also a less demanding computation that can run on most machines’ CPUs without requiring a GPU.

What about software updates – don’t these require internet access?

Yes, an internet connection is needed to download updates, but not to run the software itself. Once downloaded, updates can be distributed internally and installed offline, if needed. 

This gives organisations control over how and when they update, in line with their internal IT and data security policies.

Once installed, the software continues to operate fully offline. So, users can decide when to reconnect and update.

We do recommend that customers keep their software up-to-date. Cyclone 3DR includes monthly releases, including bug fixes and improvements to the AI engine. 

What data regulations or certifications does Cyclone 3DR comply with?

Because Cyclone 3DR runs offline and doesn’t send data to Hexagon or external servers, standard cloud compliance frameworks, like GDPR or ISO/IEC certifications, are not applicable in the same way.

The software simply doesn’t process customer data outside of the user’s environment. From a data privacy standpoint, it’s down to how the business manages their internal systems.

Will AI classification always remain a fully local process or will it move to the cloud?

AI point cloud classification will always be an offline process in Cyclone 3DR.

But in the future, AI classification will also be available in our cloud platform, Reality Cloud Studio, powered by HxDR. This will support customers with large-scale projects involving vast datasets that are larger than a local machine can efficiently handle.

Cyclone 3DR currently supports point clouds up to one billion points locally. This covers about 90% of use cases. For anything larger than this, cloud solutions will offer greater scalability and performance.

Interested in learning more about AI Classification and other Cyclone 3DR features? Browse our dedicated YouTube channel for tutorials, workflows, and tips. 

Yannick Stenger
Product Manager
Reality Capture Software