Label Studio

What Label Studio is

Label Studio is an open-source tool for labelling data for machine learning, maintained by HumanSignal. It runs in the browser and handles many data types: images (classification, bounding boxes, polygons and masks), text and documents (classification, named entities, OCR), audio (transcription, speaker labels), video (object tracking) and time series. There is a free, community-driven open-source edition and a paid Enterprise edition with extra team features.

Is it worth your time?

Yes, if your project needs labelled data you cannot download. Most student and real-world ML projects stall on exactly this. Label Studio lets you set up a labelling interface in minutes, share it with teammates, and export results in the formats training libraries expect, so you spend your time on label quality rather than on tooling.

Getting started

Install it with pip install -U label-studio and start it with label-studio; Docker and Homebrew installs are also documented. Create a project, choose a labelling template such as object detection or named entity recognition, import your images or text, and start labelling.

Export formats

The export guide lists formats including JSON, JSON_MIN, CSV, TSV, COCO, Pascal VOC XML, YOLO, CoNLL 2003, spaCy, brush masks as NumPy arrays or PNG images, and ASR manifests. COCO and YOLO cover most detection libraries; CoNLL 2003 and spaCy formats cover entity recognition.

Tips for good labels

1. Write a one-page labelling guide with examples of hard cases before you start.
2. Label a small sample twice, by two people, and compare to find unclear rules.
3. Use its model-assisted labelling to pre-fill suggestions, but check every one.
4. Keep the raw export under version control with your training code.

Free and paid

The Community edition is free and open source under Apache 2.0 and runs on your own machine or server. The Enterprise edition is paid; check the HumanSignal site for its current plans.

Common questions

Is my data private? With the Community edition running locally, your data stays on your machine.

What can I do after it? Produce clean, consistently labelled datasets for detection, segmentation, classification and entity recognition projects.

Best for: students and teams building a vision or NLP project on their own data.