Back to the articles

New image viewer for 3D labeling, attaching images to issues, and more

April 14th, 2023 - 2 min -
Avatar photo

One of the biggest challenges when labeling point clouds is locating and identifying the objects you want to label.

Is that small group of points a bike or a motorcycle?
A traffic sign or a traffic light?

No need for guessing anymore: with our new image viewer, you can always see what you’re labeling.

Synchronized image viewer for 3D point cloud labeling

Uploading camera images alongside your point clouds to provide visual context was already possible for a while.

Our new image viewer takes things a whole lot further, though: it displays a synchronized, close-up camera view of the region where you’re moving your mouse pointer in the 3D point cloud.

What is that small group of points far in the distance? Just move your mouse pointer over there and glance at the image viewer.

When an object is selected, the image viewer stays locked to it. If the object appears in multiple camera images, you can cycle through them by clicking the arrows. If you need to have a closer look at a specific camera image, click its thumbnail in the new thumbnail row to open the image full size.

Read more →

Attaching images to issues

When labeling or reviewing, you can create issues to highlight mistakes or discuss ambiguities in the labeling spec with others. This creates a tight feedback loop between labelers and reviewers to ensure errors are quickly resolved and not repeated.

You can now also attach images to issues. When creating a new issue, click the little image icon, select one or more images from your disk, or paste an image directly from your clipboard.

This new feature makes adding screenshots to your issues very easy, zooming in on specific labeling mistakes and showing how they should be corrected.

Performance improvements

We’ve optimized the performance of the point cloud cuboid interface for scenes with many cuboids. Whereas previously, the interface could become slow and laggy as the number of cuboid annotations increased, it now stays running smoothly even on older hardware.

Other features and improvements

  • When opening a camera image in the point cloud labeling interface, you can now zoom and pan to explore the image in more detail.
  • You can also toggle the point cloud overlay on and off. Disabling the point cloud overlay allows you to see the camera image in more detail while enabling it allows you to see if the camera and lidar are well calibrated.
  • In the 3D vector labeling interface, it’s now possible to add and remove points from an existing polygon or polyline.

  • A new visualization option lets you color point clouds by point height. You can now add track-level object attributes, which remain constant across all frames in a sequence.
  • Labeling metrics and timings have been moved to a separate “Insights” tab, both at the dataset level and organization level. You can select a date range and export the data.

See all changes →

Share this article