Brand new PIP package, NPM package for the embeds, and Pairwise
You’re reading the newsletter for a Label Studio release. Labeling and annotating the data is a tedious process - we know that. Last two weeks before the end of the year, we’ve decided to do a retrospective and identify how to make it less so.
Some time ago, we added the support for assisted machine learning, where you can use different models to help you annotate the dataset semi-automatically. But it was not so easy to get up and running, now it shall become much easier to create and manage new projects.
However, there is a fine line between simplicity and feature-richness. We’re still trying to identify it and would like to ask for the feedback on that release.
Simplifying the usage
With the release of the PIP package, it’s never been easier to get up and running. This code block is everything you need to execute to get the server running (note it requires >=Python3.6):
pip install label-studio
label-studio init labeling_project
label-studio start labeling_project
After the server starts, you get a new interface with support for multi-format imports and live preview of the labeling interface you’re creating.
Comparing entities in pairs to judge which is preferred, or has a higher amount of some quantitative property, or whether or not the two entities are identical. You can compare any Label Studio supported object - Texts, Images, Audios, or plain HTML between each other.
Here you can see the comparison between two HTML blocks displaying dialogues.
Separating the frontend
Along with releasing the PIP package, the frontend part moves into its own repository and a neat NPM package. If you’re looking into embedding the annotation into your application, all you have to do is
npm install label-studio
There are also docs on how to run it in development mode and extend further by adding custom annotation types. For example, videos or time series.
Till next time
We’ve postponed some features from the last release to concentrate more on ease of use, but those are in the pipeline, and we will slowly introduce it here and on our Slack channel, something that is coming:
Image Segmentation with Brushes, Livewire and Floodfill
Transformers model connector for the assisted labeling
That’s all we’ve got so far. I hope you had a great Christmas and new year celebration. Hope to hear back from you.
Till next time!
Michael, Label Studio team