Skip to content
← Models
Open weightsOpenMMLab Contributors/10 trustVerified

MMPose

The reference open toolkit for 2D pose: dozens of models, datasets, and training recipes under one Apache-2.0 roof.

Visit official page →

About

MMPose is OpenMMLab's pose estimation toolbox, and it has quietly become the reference implementation most pose papers get built on top of. It is a modular PyTorch framework covering top-down, bottom-up, and single-person pose, with configs and checkpoints for dozens of architectures across COCO, AIC, MPII, OCHuman, and whole-body benchmarks.

The newer projects/ directory is where the fast-moving work lives — RTMPose, RTMO, and RTMW all ship there. With MMDeploy export to ONNX, TensorRT, and ncnn, it spans the full path from training to edge deployment, and its benchmark tables make apples-to-apples comparison across architectures actually possible.

Specs

Context window

Image/video input (model-dependent; full COCO/WholeBody keypoint support)

Pricing

Free — Apache-2.0

Free quota

N/A (self-hosted)

API access

No hosted API — self-hosted PyTorch toolbox

Weights

Open weights

Released

Not disclosed

Benchmarks

No single published benchmark exists for the toolkit itself; it ships standardized configs that reproduce published numbers (HRNet, ViTPose, RTMPose, RTMW) on COCO, AIC, MPII, and COCO-WholeBody. Leaderboard figures belong to the individual models, not to MMPose.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

pose-model-trainingpose-benchmarkingtop-down-pose-estimationbottom-up-pose-estimationmodel-deployment

Trust score — 9/10

/10 trustVerified

Scored 9/10 (verified, metrics-v1). Evidence: OpenMMLab's flagship pose project (2020), Apache-2.0 licensed, the platform behind RTMPose/RTMO/RTMW and much of the recent pose literature; configs reproduce published results. Strengths: genuinely open, huge adoption, active development. Docked for: (1) it is a toolkit, not a model — accuracy claims belong to hosted configs; (2) the full dependency stack (mmcv, mmdeploy) can be heavy to install. License: Apache-2.0.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
10.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · OpenMMLab pose estimation toolbox and benchmark (2020); platform behind RTMPose/RTMO/RTMW.
  • · No single benchmark — hosts configs reproducing published results on COCO, AIC, MPII, COCO-WholeBody.
  • · Heavy dependency stack (mmcv/mmdeploy) is the main practical drawback.

Sources