Skip to content
← Models
Open weightsOpenMMLab (Tao Jiang, Xinchen Xie, Yining Li et al.)/10 trustMetrics-checked

RTMPose

Real-time multi-person pose built on MMPose: 75.8% AP on COCO at 90+ FPS on a CPU.

Visit official page →

About

RTMPose is OpenMMLab's answer to the gap between pose research and pose products: an empirical rethink of paradigm, backbone, keypoint localization, training strategy, and deployment inference, built on MMPose. The headline variant, RTMPose-m, reports 75.8% AP on COCO while clearing 90 FPS on an Intel i7-11700 CPU and 430 FPS on a GTX 1660 Ti.

A mobile sibling hits 72.2% AP at 70+ FPS on a Snapdragon 865, and the RTMW spin-off extends the recipe to whole-body 2D/3D pose. Everything ships under Apache-2.0 with MMDeploy backends (ONNXRuntime, TensorRT, ncnn), which is why it shows up in production pipelines that used to run OpenPose or YOLO-Pose.

Specs

Context window

Image/video input (256x192 default; 17 COCO keypoints, whole-body via RTMW)

Pricing

Free — Apache-2.0

Free quota

N/A (self-hosted)

API access

No hosted API — self-hosted weights with MMDeploy export

Weights

Open weights

Released

Not disclosed

Benchmarks

Provider-reported (project README and arXiv 2303.07399): RTMPose-m 75.8% AP on COCO with 90+ FPS (i7-11700 CPU) and 430+ FPS (GTX 1660 Ti); RTMPose-s 72.2% AP at 70+ FPS on Snapdragon 865. 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

real-time-pose-estimationmulti-person-pose-estimationedge-pose-inferencemobile-pose-estimationwhole-body-pose-estimation

Trust score — 8/10

/10 trustMetrics-checked

Scored 8/10 (metrics, metrics-v1). Evidence: technical report (arXiv 2303.07399, 2023) with detailed ablations by the OpenMMLab team; Apache-2.0 code and weights; wide real-world and TI edge-AI adoption. Strengths: open license, reproducible training configs, strong speed/accuracy balance. Docked for: (1) all speed and accuracy figures are provider-reported; (2) the flagship numbers assume tuned deployment backends. 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
8.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · OpenMMLab, arXiv 2303.07399 (2023); RTMPose-m 75.8% AP COCO at 90+ FPS (i7-11700 CPU) / 430+ FPS (GTX 1660 Ti).
  • · RTMPose-s 72.2% AP at 70+ FPS on Snapdragon 865; RTMW sibling covers whole-body 2D/3D.
  • · All speed/accuracy figures provider-reported; flagship numbers assume tuned MMDeploy backends.

Sources