Release 0.7.1

Fixed Issues:

• Empty completions from target cloud storage
• Launching ML backend from docker-compose
• Unsynchronized completion id with task id
• "undefined" popup with error

Release 0.7.0

New Features

• Connect cloud storage like s3 / gcp
• Frontend testing framework
• Putting different shapes on the same image
• maxUsages parameter for the labels
• Paragraphs

Bugfixes

• Allow different types of shapes to be used in the same image. For example you can label the same image using both rectangles and ellipses.
• Fixing double text deserialization HumanSignal/label-studio-frontend#85
• Fix bug with groups of required choices HumanSignal/label-studio-frontend#74
• Several fixes for NER labeling — empty captured text, double clicks, labels appearance

Read More

https://labelstud.io/blog/release-070-cloud-storage-enablement.html

Release 0.6.1

Post release 0.6.0 fixes & improvements:

• Remove individual task from UI #267
• More options for task sampling - enable active learning mode using --sampling=prediction-score-min
• Enabling serving locally stored images and audios by --allow-serving-local-files
• Repaired broken default localhost #287
• Fixed one-click deploy on Heroku & Google Cloud #257
• Fixed docker compose start #291
• Fixed examples for ML backend runs #294
• Fixed audio wave rendering #195
• Added HTML classification example #289
• Compatibility with Anaconda environments
• CONLL export now works fine with consecutive spans HumanSignal/label-studio-converter#4

Release 0.6.0

Nested Labeling

Nested labeling enables you to specify multiple classification options that show up after you’ve selected a connected parent class. It can match based on the selected Choice or Label value, and works with a required attribute too, smart selecting the region that you’ve not labeled. To try it out check Choices documentation and look for the following attributes: visibleWhen, whenTagName, whenLabelValue, whenChoiceValue.

Per region labeling

With per region labeling you can now provide additional attributes to the labeled regions. For example, when doing audio segmentation you can further classify the region. Per region is available for any data type and the following control tags: Choices, TextArea, and Rating.

It nicely integrates with the nested labeling, for example, you can provide multiple levels of classification for any particular region.

Machine learning backend updates

• Creating & connecting machine learning backend becomes way easier - simply define your model.py script with .fit() / .predict() methods and run ML backend with label-studio-ml start --init --script=model.py. Check quickstart and tutorials on how to connect sklearn and PyTorch models
• New ML page in UI, where you can specify URLs to connect ML backends, manually trigger model training, explore training statuses and quickly check predictions by drag-n-dropping tasks.
  
• Label Studio now supports multiple ML backends connected together. You can simultaneously get multiple predictions for each task and do comparative performance analysis for different models or different hyperparameters of a single model. It's possible to connect as many backends as you want by using --ml-backend url1 url2 ... command line option or adding them via UI.

Filtering

When the number of labels or choices is big, looking for a particular one becomes tedious. New tag to the rescue. It works with any list of Labels / Choices, and is keyboard-driven. Here is an example of the interaction. Hitting shift+f puts focus, then hitting Enter key selects the first matching item.

Display Label Names

Displaying labels on top of the labeled regions proved to be a useful feature if you’d like to do a verification of the labeling. Visually inspecting the regions takes smaller amounts of time than doing so through switching between regions.

Models Scores

Along with the names of the labels you can provide a prediction score for specific regions. That score may either come from the data that you upload or from the model that you’ve connected. When it’s available you can Sort by the score, and quickly verify/adjust the labeling for the most “uncertain” regions.

Keeping the label active

If you label the same type of data it may be cumbersome to keep selecting the same label over and over again, now you can choose to keep the last label active and use it for new labeling.

Don’t forget to unselect the region when you want to select a new label, otherwise, you’d change the label of the existing region.

Bug fixes & improvements

• --host argument now available via command-line argument (thanks to @hachreak)
• fixed upload with plain text tasks (thanks to @gauthamsuresh09)
• fixed one-click deploy on Google Cloud (thanks to @iCorv)
• fixed URL paths for proxy safety (thanks to ezavesky)

Release 0.5.1

• ability to skip tasks (the corresponding completion will be marked as "skipped": True
• add --sampling parameter for command line & display it on /tasks page
• build containerized app with docker-compose

Release 0.5.0

There are many things happened - check full release notes at the new Label Studio Blog !

Release 0.4.8

• Label config validation over connected ML backend
• Fixing an error with getting predictions when ML backend is not set

Release 0.4.7

• Video classification example
• project_name/export directory where exported results will be saved
• Fixed compatibility issues with Python 3.5 Python 3.8 and Windows (#214, #203)
• Fix orjson error (#215)
• Fix error on completion without objects or relations #197
• Actualise installation and docker deployment instructions

Release 0.4.6

A lot of bugfixes related to data import and export, please refer to #209 for details

Release 0.4.5

Major bugfixes concerning --input-path parameter on starting Label Studio: now you can really specify local paths with json-formatted files, or directory with image or audio files, e.g.:
label-studio start my_project --init --input-path path/to/local/image/dir --input-format image-dir