Jump to section
Today, we’re introducing some powerful new features that will make your life easier when segmenting large point clouds and point cloud sequences. We’re also excited to announce that we now support Gaussian splats in our 3D labeling interfaces. Let’s dive in!
Merged point cloud mode for 3D segmentation
Point cloud segmentation labeling can be incredibly time-consuming, making it one of the most expensive label types.
In the Merged Point Cloud Mode, you now annotate static elements in the scene (e.g. road surface, buildings, parked cars) in a single pass instead of having to label them in each frame individually.
Limiting cuboid feature
Segmenting point clouds consisting of millions of points can put heavy requirements on your hardware. Not everyone has a powerful GPU, sometimes causing performance issues.
With our new Limiting Cuboid feature, you can restrict your segmentation labeling to a smaller subregion of the full point cloud, alleviating performance issues.
Gaussian splat labeling
Gaussian splats are a new technique for rendering 3D scenes. They’ve gained attention for their ability to create detailed, realistic 3D scenes with relatively low computational overhead.
We’re excited to announce that you can now upload, view, and label .splat files on Segments.ai. Check out our blog post, where we show exactly how you can label your Gaussian splats.
Other features and improvements
Inside .lumen: How they scaled their annotation process
Accurate and extensive annotated data is essential for training the models that power the .lumen glasses. The team trains about ten models weekly.
Get first-hands insights on scaling and optimizing an internal annotation team, experimenting with artificual classed and ensuring quality in data annotations
Read the .lumen case study
Software Tools For Robotics Landscape (2024)
Explore the landscape of software tools for robotics in 2024. From established companies to startups, and see how they can help speed up the development of robots and autonomous vehicles.
Late vs early sensor fusion: a comparison
Multimodal fusion combines different types of data from various sources to improve prediction accuracy. Essentially, each data source provides unique and helpful information that complements the others.
There are two main approaches to multimodal fusion: late and early sensor fusion.
