Label Studio

Data annotation and labeling platform for building high-quality ML training datasets.

Overview

Label Studio is a multi-type data annotation platform for creating labeled datasets that train and evaluate machine learning models. Annotate text, images, audio, video, and time series — all in one tool, with ML-assisted pre-labeling to speed up the process. Projects, tasks, and labels are all managed through a clean web interface, and export to standard ML dataset formats.

Key Features

• Multi-Type Annotation — Text, images, audio, video, HTML, and time series in one platform
• Labeling Templates — Pre-built templates for classification, NER, object detection, sentiment, and more
• ML-Assisted Pre-labeling — Connect a model backend to auto-label tasks for human review
• Project Management — Organize annotation work into projects with task queues
• Team Collaboration — Multiple annotators work on the same project with agreement tracking
• Export Formats — Export to JSON, CSV, COCO, YOLO, Pascal VOC, and more

Getting Started

1. From the Hub, click Label Studio to launch
2. Click Create Project and give it a name
3. Upload your data (images, text files, audio, etc.) or connect a cloud storage source
4. Choose a labeling template or configure your own label schema
5. Start annotating — click through tasks and apply labels
6. Export your completed annotations when ready

Labeling Task Types

Text

• Named entity recognition (NER) — highlight and tag spans of text
• Text classification — assign categories to documents or sentences
• Relation extraction — draw links between entities

Images

• Bounding box detection — draw rectangles around objects
• Polygon segmentation — precise object outlines
• Keypoint annotation — mark specific points (e.g., body pose)
• Image classification — label the whole image

Audio

• Audio classification — label audio clips
• Transcription — type what you hear
• Speaker diarization — segment by speaker

Video

• Video classification
• Timeline segmentation

Labeling Templates

Label Studio includes templates for common ML tasks. When creating a project, choose from:

• Object Detection (YOLO/COCO format)
• Image Segmentation
• Text Classification (single/multi-label)
• Named Entity Recognition
• Sentiment Analysis
• Audio Transcription
• Time Series Annotation

Or write your own labeling config in XML to define any annotation interface.

ML-Assisted Pre-labeling

Connect an ML backend to automatically pre-label tasks:

1. Go to Settings → Machine Learning in your project
2. Add your model backend URL
3. Enable Auto-Annotation
4. Label Studio calls your model for each new task and populates predictions
5. Annotators review and correct predictions instead of labeling from scratch

Exporting Annotations

Export from the project Export tab:

When to Use Label Studio