AI / ML Computer Vision stable

Detectron2

Meta's platform for object detection and segmentation

31.0K stars 300 contributors Since 2019
Website → GitHub

Facebook AI Research platform for state-of-art object detection, instance segmentation, keypoint detection, and panoptic segmentation with extensive model zoo.

License
Apache-2.0
Min RAM
4 GB
Min CPUs
2 cores
Scaling
single_node
Complexity
advanced
Performance
enterprise grade
Self-hostable
K8s native
Offline
Pricing
fully free
Docs quality
good
Vendor lock-in
none

Use cases

  • Train custom object detection models for manufacturing QA
  • Instance segmentation for autonomous driving
  • Keypoint detection for pose estimation
  • Panoptic segmentation for scene understanding

Anti-patterns / when NOT to use

  • PyTorch only — no TensorFlow support
  • Research-focused — production deployment needs extra work
  • GPU required for reasonable performance
  • Documentation assumes ML expertise

Compare with alternatives

Replaces / alternatives to

  • Google Vision API
  • AWS Rekognition

Technical specs

Language
Python
API type
SDK
Protocols
HTTP
Deployment
pip
SDKs
python

Community

GitHub stars 31.0K
Contributors 300
Commit frequency monthly
Plugin ecosystem none
Backing Meta FAIR
Funding corporate

Release

Latest version
Last release
Since 2019

Best fit

Team size
solosmallmedium
Industries
automotivesecuritymanufacturinghealthcareretail

Tags

  • object-detection
  • segmentation
  • keypoints
  • panoptic
  • model-zoo
  • mask-rcnn
  • retinanet