Labelbox December updates

Wrapping up 2020 and heading into the New Year, we're happy to announce a round of recent and upcoming improvements that make Labelbox faster, more powerful, and more secure.

Model-assisted labeling expands to video, text, and nested classifications.

In October we announced model-assisted labeling, a powerful way to use your own model to accelerate labeling and model performance that allowed customers to reduce labeling time by 50-70 percent. We’re now excited to expand model-assisted labeling to video frame classification, text entity annotations, and nested classifications for images.

Labelbox’s video tool was already engineered for speed, but with model-assisted labeling, it’s now even faster. Import predictions from your own model or a model of your choice to automate Radio or Checklist frame classifications. Model-assisted labeling helps labelers focus their time where it’s needed the most. For example with video, model predictions can guide labelers to frames where specific objects might be present so a labeler can get straight to annotating rather than combing through hundreds or thousands of frames.

In early January, model-assisted labeling will go a layer deeper on image annotations and be able to import model predictions for all nested classification annotation types. With this update customers will be able to pre-label data with more fidelity and apply predictions to more complex ontologies.

We’ve been thrilled to watch customers incorporate model-assisted labeling into their workflows and with this round of updates are confident that more customers will be able to further accelerate their model performance improvements. Stay tuned for even more improvements to model-assisted labeling in Q1, and head over to our documentation to learn how to get started.

Improved data security for AWS storage

By the end of January, customers storing data in Amazon S3 will be able to use delegated access to integrate with Labelbox more simply and securely than ever. Delegated access gives Labelbox read-only access to pre-process data, ensuring quick and reliable load times for labeling tools and full access to the latest Labelbox features.

Historically, customers with more stringent security requirements using their own S3 storage without public URLs have had to apply a blanket IP restriction, regenerate signed URLs, or deal with a Heroku proxy in order to add a layer of security. Because Labelbox needs read access to pre-process assets, this hybrid cloud configuration hasn’t supported some of our advanced tools like superpixel and the video tool.

If you’re interested in setting up delegated access for your Amazon S3 storage, sign up to be a beta user here. Support for other cloud storage providers will be added later in the year.

Annotation-based usage analytics

We’re rolling out new usage analytics for tracking annotation volume across image, video, and text data rows. This will give customers more detailed usage insight than is available when just looking at label volume, and will be available in the Labelbox application user interface and via the API. Annotation-based analytics will be available in January, 2021.

Add up to 10 image overlay layers

The image editor now supports up to 10 image overlays so you can add even more context to your images and improve labeling accuracy on complex images. Switch between layers with hotkeys so you can quickly jump to the most informative view.


From the entire Labelbox team, we wish you a happy New Year! We’re looking forward to 2021 and to a host of upcoming feature updates and releases, and can’t wait to share them with you. Until then, happy holidays!

Tags

Labelbox

Labelbox is a collaborative training data platform empowering teams to rapidly build artificial intelligence applications.