autodistill/autodistill

Health v2 · maintenance only

Adoption not part of the score

2763 stars · 221 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Autodistill is a Python library that uses large foundation vision models (like Grounding DINO, Grounded SAM, and CLIP) to automatically label unlabeled image datasets, then trains small supervised models (like YOLOv5, YOLOv8, and DETR) on those auto-labeled datasets. It provides a pluggable pipeline that goes from raw images to a fine-tuned, deployable model with no human labeling in between.

Use cases

• auto-label images for object detection without manual annotation
• train a custom YOLOv8 model from unlabeled photos
• distill a large vision foundation model into a small fast model for edge deployment
• generate labeled datasets using Grounding DINO or Grounded SAM
• build an end-to-end image labeling and model training pipeline with no human intervention
• create instance segmentation training data automatically
• label a dataset with text-prompted classes using an ontology

When to choose

• you have many unlabeled images and want to avoid manual annotation
• you need a small, fast supervised vision model distilled from a foundation model
• you want to deploy custom object detection or segmentation models to edge devices
• you want a pluggable pipeline connecting base models like Grounded SAM to target models like YOLOv8

When to avoid

• you need highly accurate labels that require human review and verification
• your task is not object detection or instance segmentation (classification support is still limited)
• you lack GPU hardware or cannot run large foundation models locally
• you need a fully managed labeling service rather than a self-run pipeline

Facets

library · maturity active

machine-learning computer-vision image-processing data-generation deep-learning machine-learning computer-vision image-processing deep-learning artificial-intelligence python cross-platform model-distillation auto-labeling foundation-models object-detection instance-segmentation image-annotation grounding-dino segment-anything yolov8 pytorch dataset-labeling edge-deployment gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="autodistill/autodistill")