Back to the articles

Text labeling, dataset cloning, changelog and more

June 16th, 2022 - 1 min -
Avatar photo
-

At Segments.ai we’re focused on computer vision, but sometimes our customers come to us with use cases where text labeling is also needed. We’re happy to announce that you can now label text directly in Segments.ai, without having to switch to another tool.

Text labeling

To support the labeling needs of our users beyond computer vision, we’re launching two new text-labeling interfaces:

  • The Named Entity Recognition (NER) interface can be used to tag words and phrases in a text snippet with non-overlapping labels
  • The Span Categorization interface is useful when working with overlapping categories.

Upload your text dataset through the web platform or Python SDK to get started. Label the text with your mouse or by using the (configurable) hotkeys. When finished, create a release and download your labeled dataset.

Try it out and let us know your thoughts; all feedback is welcome!

Cloning datasets

Another often-requested feature we’re launching is the ability to clone datasets.

Cloning a dataset is useful when you want to label the same data with different label types or categories. For example, you’ve already labeled all objects in your dataset with bounding box labels and now you’d like to label the road, sky and other regions with segmentation masks. This is now very easy: just clone the dataset, change the task type and categories, and start labeling.

For new users, cloning one of our public example datasets is also a great way to try out our labeling interfaces without having to upload any data. From your home screen, click the “New dataset” dropdown icon and select “Clone an example dataset.” We have example datasets for each sample and label type.

Changelog

Make sure to check out our new changelog feature in the top navigation bar. It keeps you up to date about our latest features, improvements and bug fixes.

Other features and improvements

  • In the point cloud cuboid interface, you can toggle an info pane to inspect the dimensions of the selected cuboid.

  • Cuboid rotation is made more intuitive and can now also be done using hotkeys.
  • In the image and point cloud labeling interfaces, you can now see the object counts per category by hovering over the little pie diagram icon in the objects sidebar.
  • The design of the user overview and settings page is improved.
  • You can now export image vector datasets to COCO format in the Python SDK.
  • The Autosegment feature now also works on 4-channel images.

See all changes →

Share this article