OpenPose
The bottom-up multi-person pose library that defined the field: body, face, hands, and feet keypoints from one network.
About
OpenPose was the first widely usable library for real-time multi-person 2D pose estimation, and it is still the one every later whole-body system gets compared against. Its trick is Part Affinity Fields: instead of detecting people and then finding their joints, it detects every joint and limb in the image at once and assembles the people from the bottom up, so runtime barely changes as crowds get bigger.
The library bundles body, face, hand, and foot keypoint models (135 keypoints in the whole-body variant) in a C++ core with Python, Unity, and Unreal wrappers. ControlNet made it the default pose preprocessor for image generation, though distilled successors like DWPose have since replaced it in most new pipelines. The models are free for non-commercial use; commercial use needs a CMU FlintBox license.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-claimed: the Part Affinity Fields CVPR 2017 paper reported 61.8 mAP on COCO test-dev keypoint detection. The library itself was never published with benchmark numbers as a product, and no independent benchmark of its packaged models was found.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified, metrics-v1). Evidence: peer-reviewed CVPR 2017 Part Affinity Fields paper and TPAMI 2019 OpenPose paper by named CMU researchers; among the most-cited and most-deployed pose systems ever; code and models genuinely downloadable. Strengths: foundational provenance, full whole-body keypoints, enormous ecosystem (ControlNet, research tooling). Docked for: (1) weights are non-commercial licensed, not truly open; (2) maintenance stalled since ~2020 and accuracy is now beaten by successors. License: custom OpenPose license (non-commercial); commercial via CMU FlintBox.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 10.0/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 6.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · CVPR 2017 Part Affinity Fields paper + TPAMI 2019 OpenPose paper by CMU PCL (Cao, Hidalgo, Simon, Wei, Sheikh).
- · Provider-claimed 61.8 mAP on COCO test-dev (PAF paper); library never independently benchmarked as a product.
- · Repo maintenance stalled since ~2020; still the reference whole-body baseline and ControlNet's original pose preprocessor.