Company NewsProcessing & Storage

Pointly’s new AI training service tailors point cloud classifications to client datasets

Press Release 30 Mar - 13:22 SAST

Pointly’s new AI training service tailors point cloud classifications to client datasets

Press Release 30 Mar - 13:22 SAST

Information from Pointly

Pointly’s new AI training service offers clients automated and customer-specific classifications of point clouds as well as tailor-made neural networks that are trained and tested for a client’s data.

The service includes:

  • A Professional Pointly subscription to upload, manage and classify point clouds.
  • A personal consultation session with the company’s experts to assess your goals and define KPIs.
  • Regular performance updates on the training progress of the model.
  • An AI point cloud classifier (i.e. the neural network classifying the data).

After the service ends, clients can keep using their AI point classifier for classification tasks as well as re-train it with more data based on standard charges.

The service requires training data for Pointly’s data scientists to train the model. To generate this data, the point clouds need to be uploaded into Pointly and manually classified – either using Pointly’s selection tools or using a labelling service provider from the company’s partner network.

Approximately 5 to 15 days of labelling work tends to be suitable for a proof of concept of the future potentials of AI. An iterative training approach splits the data into three evenly sized sets, beginning training with the first, then adding the second and so forth. This allows the company to evaluate the model’s performance and gives customers more flexibility and control when generating training data.

Customer can choose a test site that will not be used for training but only for evaluating the final performance of the neural network.

After the initial training, the model can take over classifying the point clouds automatically. Users can still manually adjust the model outputs inside Pointly to improve the model, since repeated improvements make the model more robust.

Find out more.