HumanSignal/label-studio-ml-backend

What it solves

Label Studio ML backend is an SDK that allows developers to wrap machine learning models and deploy them as web servers. This enables the automation of labeling tasks within a Label Studio instance by providing pre-annotations, interactive predictions, and the ability to train models on submitted annotations.

How it works

The SDK provides a base class, LabelStudioMLBase, which developers can inherit from to implement their own inference and training logic. By overriding the predict method, the server can send predictions back to Label Studio. By overriding the fit method, the server can update the model state or save weights based on new annotations.

Who it’s for

Machine learning engineers and data scientists who use Label Studio for data labeling and want to integrate their own custom models to speed up the labeling process through automation.

Highlights

• Pre-annotation: Automatically generates initial labels for data before a human reviewer sees it.
• Interactive Mode: Provides real-time predictions as the user performs actions on the labeling page.
• Training Integration: Allows models to be updated or retrained based on the labels provided by humans.
• Extensive Model Examples: Includes pre-built examples for a wide range of tasks including BERT classifiers, SAM 2 image segmentation, YOLO object detection, and RAG pipelines via LangChain.
• Deployment Options: Supports deployment via Docker, GCP, and Nebius Serverless.