3D point-cloud segmentation (experimental)
Assign semantic labels to individual points in PLY and PCD files on your device. This experimental workspace runs in a desktop browser or desktop app with WebGL2. It labels points, not 3D cuboids, and is unavailable in native mobile apps and the hosted API service.
Create and label a cloud
- Create a project using the 3D Point Cloud card and add your class labels. The Experimental badge describes the current scope. You can also open the 3D Scans sample project to explore the workspace.
- Import PLY or PCD files. Each cloud can contain up to 5 million source points; parser memory and file-size limits also apply. PLY mesh faces are ignored: this workspace edits the points.
- Open a cloud, orbit or pan, and choose Perspective, Top, Front or Side. Adjust point size and RGB, intensity or class colouring when the data supports it.
- Use Lasso or Brush to select points. Visible-surface mode picks the front layer; Through includes points behind it. Shift adds to the selection and Alt removes from it. Clipping and class visibility help isolate a region.
- Choose a class and Assign, or Unlabel the selected points. Undo and Redo work within the current editing session. Save before closing; reopen the cloud to continue from the saved labels.
Formats and portability
PLY and PCD are accepted; LAS/LAZ and general mesh editing are not supported. Invalid coordinates keep their source positions but cannot be selected. Each saved class label stays aligned with the original point order.
Export point labels to a ZIP containing source clouds, the class mapping and per-point labels. This bundle is an export for downstream use, not a label-import workflow. To reopen the complete annotated project in AnnotateIt, use Export project and import that project archive, or restore a full profile backup.
Current limits
- One cloud at a time. No 3D cuboids, sequences, sensor poses, sensor fusion or multi-cloud scenes.
- Manual screen-space lasso and brush selection; no region growing or AI-assisted 3D annotation.
- No dataset versions, splits, quality scans, model evaluation or 2D dataset-format exports for 3D projects.
- No conversion between 2D and 3D projects. Native iOS/Android and browsers without WebGL2 cannot edit point clouds.
- The current source includes the workspace; installed store builds may lag behind. Browser validation does not establish performance on every packaged desktop runtime.