v0.8.2
HumanSignal/label-studiov0.8.2Dec 1, 2020by niklub
AI Summary
Label Studio v0.8.2 introduces significant new labeling capabilities including Paragraphs for structured text with audio pre-listening, Taxonomy for nested classification trees, and improved undo/redo functionality. This release also adds project sharing capabilities, new UI settings like line numbers, and various ML backend enhancements.
Key Highlights
- Paragraphs - New labeling format for structured JSON text with audio pre-listening capability using start/end timestamps
- Taxonomy - New control tag wrapper for building nested multi-class classification trees
- Undo/Redo/Reset - Enhanced history tracking with visual state indicators and full keyboard shortcut support (cmd+z/ctrl+z)
- Project sharing - Ability to move and share entire project folders across different paths
- New ML backend for chatbot responses powered by HuggingFace's GPT2
New Features
- Paragraphs labeling for structured text with author/text keys and optional audio attachment
- Taxonomy control tag for nested choice hierarchies
- Undo/Redo/Reset with action history state visibility
- TextArea regions with multiline support (shift+enter), escape to cancel, and trash icon deletion
- Line numbers for Text/Paragraphs
- Select region after creation for better per-region labeling
- Project local path accessible on ML backend API
- New ML backend for chatbot responses from Paragraphs (HuggingFace GPT2)
- Move and share entire project folders
Full Release Notes
## Paragraphs
Now you can label structured text in simple json format. Introducing `Paragraphs`!
It requires array of phrases, defined as object with two required fields:
first one for author (set by `nameKey`, "author" by default) and second one for text itself
(set by `textKey`, "text" by default). For better appearance use `layout="dialogue"`.
#### Audio pre-listening
Also you can attach audio for these phrases by `audioUrl`.
It uses optional fields in phrases' objects: `start` and `end`/`duration` in seconds,
linking phrases with according part of audio.
<div style="margin:auto; text-align:center; width:100%"><img src="https://labelstudioimg.s3.amazonaws.com/label-studio-paragraphs-with-audio.png"/></div>
## Taxonomy
`<Taxonomy>` is a new control tag wrapper that allows you to build a nested choices structure - a multi-class classification tree. The usage is quite intuitive:
```xml
<Taxonomy name="" toName="text">
<Choice value="Archaea" />
<Choice value="Bacteria" />
<Choice value="Eukarya">
<Choice value="Human" />
<Choice value="Oppossum" />
<Choice value="Extraterrestial" />
</Choice>
</Taxonomy>
```
Learn more by playing with _Text taxonomy_ template on **Setup** page!
## Undo/Redo/Reset
Now you can see the current state of your actions history — is it possible to undo/redo.
Also reset can be undone.
Both `cmd+z` and `ctrl+z` are working now.
## TextArea regions
Improve text comments, especially multiline ones — they appears much better than before,
you can add them by `shift+enter` or cancel editing by `escape`.
Also you can delete classification comments with trash bin icon next to it.
And they will back to Regions list soon.
## Machine Learning Backend Features
* New ML backend for chatbot responses generation from `<Paragraphs>` (powered by [HuggingFace's GPT2](https://huggingface.co/models?filter=gpt2))
* Project local path is accessible on ML backend API - in case you want to read uploaded images/audio directly from disk
## New UI settings
* Line numbers for Text/Paragraphs
* Select region after create for better per-region labeling
## Project sharing
Now it's possible to move and share the entire project folder and start annotations from any other path
## Fixes
* Restore region's text for NER labeling from data if it was missed
* Display spinner while loading heavy TimeSeries
* Return back `onEntityDelete`/`onEntityCreate` callbacks
* Various fixes Image regions stability
* Performance improvements for large regions numbers
* Task IDs are now safely restored, even if you break or delete `source.json`
* Get rid of some packages dependencies error
* Export to VOC with broken image paths
* Windows path for ML backend https://github.com/heartexlabs/label-studio/issues/500
* Export to specific formats with relation results