Data and AI tooling
LiDAR classification automation, PointNet++, graph neural networks and result control on real data.
We combine engineering experience with data and AI tooling: PointNet++, graph neural networks and classification automation.
Models accelerate the first processing pass. We inspect scene geometry, rare classes and object boundaries; final accountability for classification stays with a specialist.
The task
Reduce repetitive manual geodata processing and make results reproducible while retaining engineering control where a model encounters rare classes or ambiguous geometry.
Input data
Scoping requires representative data, a description of target classes and the expected result; training also requires suitable verified labels. We assess availability and quality before selecting a method.
Workflow
- Define a measurable task and select data for evaluating the approach.
- Prepare the pipeline, train or configure the model and evaluate it on a separate sample.
- Analyse errors, refine the process and hand over the tool with a clear control procedure.
Deliverables
Depending on the task, the result may be a configured data-preparation or classification workflow, a verified output set and a description of operating limits. The exact handover format is agreed for each project.
Quality control
We evaluate results on data not used for configuration and separately inspect rare classes, object boundaries and difficult scenes. Model output is not accepted without the specialist control defined for the project.