DWPose
Whole-body pose by two-stage distillation: the model that replaced OpenPose as ControlNet's preprocessor.
About
DWPose distills a heavyweight whole-body teacher into a family of fast students — tiny through large — that predict all 133 COCO-WholeBody keypoints (body, hands, face, feet) in one pass. A two-stage distillation strategy lifts the distilled RTMPose-l from 64.8% to 66.5% whole-body AP, even beating the larger teacher.
Its real fame is downstream: it replaced OpenPose as the default pose preprocessor in ControlNet, and its clean ONNX export (the default branch) makes it the pragmatic choice for generation pipelines, character animation, and anywhere whole-body skeletons feed a diffusion model.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported (ICCV 2023 workshop paper): 66.5% whole-body AP on COCO-WholeBody, up from the 64.8% RTMPose-l teacher and above the 65.3% RTMPose-x teacher. No independent published benchmarks were 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 ICCV 2023 workshop paper by named IDEA/Tsinghua researchers; Apache-2.0 code and weights; ~2,800 GitHub stars and adoption as ControlNet's default pose preprocessor. Strengths: open license, real adoption in production pipelines, reproducible distillation recipe. Docked for: (1) workshop-level peer review, not a main-conference paper; (2) benchmark figures are provider-reported. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · ICCV 2023 workshop (Yang, Zeng, Yuan, Li; IDEA/Tsinghua SIGS); two-stage distillation of RTMPose.
- · Provider-claimed 66.5% whole-body AP on COCO-WholeBody (beats RTMPose-x teacher at 65.3%); ~2,800 GitHub stars.
- · Replaced OpenPose as ControlNet's default pose preprocessor; ONNX branch is default.