/
Automatic Classification

Automatic Classification

Menu: AI / Automatic Classification

Version française

With this feature, you can automatically classify your projects using a trained AI model.

 

image-20260107-172335.png

Engine Selection

You must choose the engine used for training:

VisionLidar 1.3

  • Legacy engine based on CNNs

VisionLidar 2.0 (Recommended)

  • More than 10× faster

  • More than 10× more GPU-memory efficient

  • Higher model accuracy

  • Can classify billions of points on an average PC or laptop in a matter of minutes.

Selecting a Model

The software provides four default models, each designed for different dataset types. If you have previously trained a custom model, it will appear in the "Model Name" dropdown menu. Once a model is selected, the available classification classes will be displayed.

Classification Modes

You can choose between the following processing options:

🔹Performance Mode
Adjusts the batch size and number of workers based on available GPU memory:

  • Minimal → For GPUs with less than 12GB memory.

  • Balanced → For GPUs with 12–16GB memory.

  • High → For GPUs with 24GB or more memory.

Choosing the correct performance mode ensures optimal memory usage and prevents crashes due to insufficient GPU resources.

🔹 By Corridor Mode (For Linear Projects)

  • Classifies only the points inside a defined corridor.

  • Corridors can be created using the Alignments tool.

  • Best suited for road, railway, or pipeline projects where classification should be restricted to a predefined area.

🔹 Ignoring a Class (For VisionLidar 2.0 only)

  • Allows users to exclude selected classes during the Automatic Classification process without retraining the model.

  • Points that would normally be classified under an ignored class are reassigned to another class depending on model confidence.

  • The option applies only at inference time and does not modify the trained model.

  • The same model can be reused with different ignored-class configurations for different projects or workflows.

Starting the Classification

Once all parameters are set, click [Start] to begin the automatic classification process.

By selecting the appropriate mode and settings, you can optimize the classification process for your dataset and hardware. 🚀

Log Tab

The Log tab records:

  • Training steps

  • Warnings

  • Errors and failures

If the classification fails or behaves unexpectedly:

  • Send the log file to the VisionLidar support team

  • Logs allow rapid debugging and precise issue resolution