Label Studio ist die führende Open-Source Plattform zum Annotieren von Trainingsdaten für ML-Modelle.

Installation

pip install label-studio
label-studio

Browser: http://localhost:8080

Features

  • Multi-Format: Text, Images, Audio, Video, PDFs
  • ML-Assisted: Auto-labeling mit vortrainierten Modellen
  • Active Learning: Prioritize hard examples
  • Team Collaboration: Multi-user projects
  • API: REST API für Automation

Labeling Tasks

Text Classification

Wähle einer oder mehrere Kategorien pro Text.

Named Entity Recognition

Markiere Entitäten (Person, Ort, Organisation).

Image Bounding Box

Zeichne Boxen um Objekte.

Audio Transcription

Transkribiere Sprache zu Text.

Workflow

1. Create Project
2. Upload Data (CSV, JSON, Images)
3. Define Label Schema (Kategorien)
4. Annotate (Team oder ML-Assisted)
5. Export (COCOformat, CSV, JSON)
6. Train Model

API Example

from label_studio_sdk import Client

ls = Client(url='http://localhost:8080', api_key='your-key')
project = ls.create_project(
    title='My Project',
    label_config='<View>...</View>'
)

# Upload data
project.import_tasks([
    {'data': {'text': 'Sample text'}}
])

Sources