What does LumiScan Training produce?

A neural network that finds your objects – which object, where and at which angle – saved as an ONNX file that LumiScan AI Object Matching and LumiScan VGR load. With the optional Hailo PlugIn it is also converted for a Hailo TPU.

How many training images are needed?

In many applications, 10 to 20 annotated images are enough for robust results. The images are annotated with LumiScan Annotation.

What are the system requirements?

Windows 11 Professional or Enterprise, an NVIDIA RTX graphics card with at least 16 GB of graphics memory (24 GB recommended), at least 16 GB of system memory (32 GB recommended) and a licence on the LumiScan Key.

Do we need Python, CUDA or a cloud service?

No. Everything is included in the installer; the computer only needs an NVIDIA graphics driver. Training runs entirely on-premise.

LumiScan Ecosystem: LumiScan Engineering

LumiScan Training Framework

Train high-accuracy detection models from as few as 10–20 images.

Annotation and Training Program for Object Detection

The annotation and training program supports the precise annotation of image data for object detection. An integrated annotation tool enables the efficient creation of the required data formats and ensures high data quality.

Based on this data, a specially optimized training pipeline allows the development of high-performance detection models even with very small datasets. In many use cases, 10 to 20 annotated images are sufficient to achieve robust, economical, and reliable results.

The entire toolset is fully deployable on-premise and operates without any cloud connection. This ensures the user retains complete data sovereignty at all times.

All apps in the ecosystem

LumiScan Training Framework

Frequently asked questions

Interested in LumiScan Training Framework?

Let’s talk about your application.