Point Cloud Custom Classifier.

Your custom-made automatic point cloud classifier – trained and tested for your use case.

What Your Custom AI Model for Point Clouds Can Do.

Manually annotating 3D point clouds takes time.

With a custom-trained AI model, you can significantly reduce that effort. Classification becomes automated and tailored to your specific use case.

We develop a model that fits your requirements exactly, helping you increase efficiency and unlock new possibilities. Some customers report that classification time in certain projects dropped from several days (with manual work) to just a few hours.

a graph showing the advantages of the AI Training Service over manual 3d point cloud annotation

How do Point Cloud Custom Classifiers work?

You can either quickly generate training data yourself in our Pointly platform or through our labeling workforce. If you have already classified data, you can leverage it as training material for your classifier and get a jump start with our 3D Point Cloud Services.

Phase 1 – Creating Training Data

To train a high-accuracy custom AI model, high-quality training data is essential as input for the neural network.

You can easily create training data yourself using the Pointly platform, or assign the task to our dedicated labeling team.

If you already have pre-classified datasets, they can also be used as training material.

Phase 2 - Training the AI Model

Once the training data is prepared, Pointly’s 3D AI specialists begin the iterative training process. In addition to deep learning, we develop a custom pre- and post-processing pipeline based on your project requirements. This may include rule-based methods to further refine the AI's classification output.

The result is an AI model that can fully automate the classification of your point clouds.

Phase 3 - Using Your AI Model

Once your model has been trained and deployed to the Pointly platform, there are two ways to use it for automated point cloud classification:

Option 1: The Pointly Platform

 

Your custom AI model will be available in the upload window of your Pointly account. When uploading a new dataset, you simply select your model during the classification step.

Using the platform allows you to interact with your point clouds, review and edit classification results, share projects with others, and generate vector data. It also serves as a full-featured management system for point clouds, offering flexible storage options.

Option 2: The Pointly API

This is a fully automated classification service, used via API with no manual interaction on the platform. Your data is sent to the selected model via API, and the classified files are automatically returned.

The benefit: seamless integration into your existing workflows – simple, fast, and scalable.
The trade-off: manual tools and the visual interface of the Pointly platform are not included in this option. It is intended purely for automated processing.

Do you want to know more, or do you have any questions? Please read our FAQ or send us an email to sales@pointly.ai!

Point Cloud Custom Classifier FAQ.

Simply tell us about your use case in the linked form. It takes less than 3 minutes! Or send us an email to sales@pointly.ai

Once we have trained a neural network classifier for your use case, we can automatically classify new cloud data.

  • We will provide you with a Pointly Professional License. That means you can upload your point clouds into a custom project.
  • Does your data already contain labels? Great, we can start straight away with training a classifier on those.
  • Don’t worry: Pointly was built for quickly generating training data if you don’t have labeled data. You can either do it yourself or order a labeling service through us.

5-15 days of labeling work provide a realistic start for a Proof of Concept. That means you will get a solid estimate of the future potentials of AI but not yet a production-ready model. This number can, of course, vary a lot depending on your use case and the desired performance. We are offering you a briefing call where we clarify all specifics that apply to your use case. You can choose to either invest that effort yourself or hire a labeling service from us at competitive rates. Naturally, more training data is always beneficial for training the AI model.

The first and most crucial step is that you choose a test site to be put aside. We will not train on this test site but will use it to evaluate the final performance of the neural network. We will keep you updated on the current performance: based on cross-validations within the current set during the training process on the separate groups of data.

Split the available data into three evenly sized sets, beginning with the first, then adding the second, and so forth. Why this iterative approach?

  • It allows us to evaluate how performance increases with additional training data
  • It gives you more flexibility and control when generating training data.

After every iteration, we provide you a short report on the current performance.

Afterward, you can keep using your neural network for classification tasks and re-train it with more data based on standard charges.

Please fill out the form down below to inquire about an AI Training Service with us.

 


You are currently viewing a placeholder content from HubSpot. To access the actual content, click the button below. Please note that doing so will share data with third-party providers.

Unblock content Accept required service and unblock content
More Information

Pointly developer icon

Developed by Experts.

Geo-Data Science Technology made in Germany