Label Studio: Data Labeling, Annotation & ML Integration

Setup & Project Configuration

Run Label Studio locally or via Docker, then configure labeling projects with XML templates:

# Install and run (local)
pip install label-studio
label-studio start         # opens http://localhost:8080

# Or Docker
docker run -it -p 8080:8080 \
  -v $(pwd)/mydata:/label-studio/data \
  -e LABEL_STUDIO_LOCAL_FILES_SERVING_ENABLED=true \
  heartexlabs/label-studio:latest

# Create a project via Python SDK
from label_studio_sdk import Client

ls = Client(url="http://localhost:8080", api_key="your-api-key")
project = ls.create_project(
    title="Product Image Classification",
    label_config="""
<View>
  <Image name="image" value="$image"/>
  <Choices name="category" toName="image" choice="single">
    <Choice value="Electronics"/>
    <Choice value="Clothing"/>
    <Choice value="Furniture"/>
    <Choice value="Food"/>
  </Choices>
</View>
    """
)

# Import tasks (URLs or local files)
project.import_tasks([
    {"image": "https://example.com/img1.jpg"},
    {"image": "https://example.com/img2.jpg"},
])

NER & Bounding Box Labeling Configs

Label Studio uses XML-based configs to define labeling interfaces for any task type:

<!-- Named Entity Recognition (NER) config -->
<View>
  <Labels name="label" toName="text">
    <Label value="PERSON" background="blue"/>
    <Label value="ORGANIZATION" background="green"/>
    <Label value="LOCATION" background="orange"/>
    <Label value="DATE" background="purple"/>
    <Label value="MONEY" background="red"/>
  </Labels>
  <Text name="text" value="$text"/>
</View>

<!-- Object Detection (bounding box) config -->
<View>
  <RectangleLabels name="label" toName="image">
    <Label value="Car" background="#FF0000"/>
    <Label value="Person" background="#00FF00"/>
    <Label value="Bicycle" background="#0000FF"/>
  </RectangleLabels>
  <Image name="image" value="$image" zoom="true" zoomControl="true"/>
</View>

<!-- Polygon segmentation config -->
<View>
  <PolygonLabels name="label" toName="image" strokeWidth="3">
    <Label value="Road" background="#808080"/>
    <Label value="Building" background="#A0522D"/>
    <Label value="Vegetation" background="#228B22"/>
  </PolygonLabels>
  <Image name="image" value="$image"/>
</View>

ML-Assisted Labeling with Active Learning

Connect a prediction server to pre-annotate tasks and close the active learning loop:

# label_studio_ml backend (FastAPI-based prediction server)
from label_studio_ml.model import LabelStudioMLBase

class NERPredictor(LabelStudioMLBase):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        import spacy
        self.nlp = spacy.load("en_core_web_lg")

    def predict(self, tasks, **kwargs):
        predictions = []
        for task in tasks:
            text = task["data"]["text"]
            doc = self.nlp(text)
            results = []
            for ent in doc.ents:
                results.append({
                    "from_name": "label",
                    "to_name": "text",
                    "type": "labels",
                    "value": {
                        "start": ent.start_char,
                        "end": ent.end_char,
                        "text": ent.text,
                        "labels": [ent.label_],
                    },
                    "score": 0.9,
                })
            predictions.append({"result": results, "score": 0.9})
        return predictions

# Run: label-studio-ml start . --port 9090
# Connect in Label Studio: Settings → ML → Add ML Backend → http://localhost:9090

Label Studio vs Scale AI vs V7 vs CVAT