Detectron2
MetaMeta AI's library for object detection, segmentation and other visual recognition tasks, the successor to Detectron and maskrcnn-benchmark. It implements Faster and Mask R-CNN, RetinaNet, keypoint, panoptic and DensePose models among others, with a model zoo of COCO, LVIS and Cityscapes baselines and training tools that many research projects build on. It installs from source or from Linux wheels Meta hosts, not from PyPI.
Openness
5 high confidence- license
- Apache-2.0(OSI
- source
- public(github.com/facebookresearch/detectron2)
- core features withheld
- no — Meta research library
- model-zoo-terms
- CC-BY-SA-3.0(optional model-zoo weights
The code is Apache-2.0 and builds from the public repository; Meta sells nothing around it. The model-zoo weights are licensed separately under Creative Commons Attribution-ShareAlike 3.0. The last tagged release was v0.6 in November 2021, though the repository is still updated.
- https://raw.githubusercontent.com/facebookresearch/detectron2/main/LICENSE recorded 2026-09-27
LICENSE is the Apache License, Version 2.0.
- https://raw.githubusercontent.com/facebookresearch/detectron2/main/MODEL_ZOO.md recorded 2026-09-27
MODEL_ZOO: "All models available for download through this document are licensed under the Creative Commons Attribution-ShareAlike 3.0 license."
- https://raw.githubusercontent.com/facebookresearch/detectron2/main/README.md recorded 2026-09-27
README: "Detectron2 is Facebook AI Research's next generation library that provides state-of-the-art detection and segmentation algorithms."; no paid tier mentioned.
- https://ungh.cc/repos/facebookresearch/detectron2/files/main recorded 2026-09-27
Repository tree with no ee, enterprise, commercial or pro directory; training scripts tools/train_net.py, plain_train_net.py and lazyconfig_train_net.py.
- https://ungh.cc/repos/facebookresearch/detectron2/releases recorded 2026-09-27
Latest release v0.6, published 2021-11-15.
Adoption
3 low confidenceGitHub stars are the only usable signal, because Detectron2 is not published on PyPI - the name is registered with no files - and installs from source or from Meta's own wheel host, which publishes no counts. A star is not a use, and the stars scale stops at 3.
- https://detectron2.readthedocs.io/en/latest/tutorials/install.html recorded 2026-09-27
Install guide: build from source with "python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'", or "Install Pre-Built Detectron2 (Linux only)".
- https://pypi.org/simple/detectron2/ recorded 2026-09-27
The PyPI simple index for detectron2 lists no distribution files.
- https://ungh.cc/repos/facebookresearch/detectron2 recorded 2026-09-27
Repository record for facebookresearch/detectron2: 34,734 stars, last pushed 2026-08-19.
Capability
3 high confidenceDetectron2 is a model zoo with the training framework around it, for detection and segmentation where timm covers image backbones. Its baselines predict the benchmark classes they were trained on, so a new label set needs a training run.
- https://raw.githubusercontent.com/facebookresearch/detectron2/main/MODEL_ZOO.md recorded 2026-09-27
MODEL_ZOO: "a large collection of baselines trained with detectron2 in Sep-Oct, 2019"; sections for COCO object detection, instance segmentation, person keypoint detection, panoptic segmentation, LVIS, Cityscapes and Pascal VOC baselines.
Verified 2026-09-27