Skip to content
← Models
Open weightsGoogle (TensorFlow.js team; based on PersonLab by George Papandreou, Tyler Zhu et al.)/10 trustMetrics-checked

PoseNet

The browser pose model that started it all: real-time human pose in TensorFlow.js from 2018.

Retired

PoseNet appears to be discontinued or shut down. This page is kept as an archive record and is excluded from search engines.

Visit official page →

About

PoseNet was Google's first public pose model for the browser, and for a while it was the default answer to 'how do I do pose estimation in a webpage.' It shipped as a TensorFlow.js package estimating 17 COCO keypoints per person, with single-person mode and a multi-person mode driven by the bottom-up greedy decoding from Google's PersonLab paper.

Model choices (MobileNetV1 for speed, ResNet50 for accuracy) and knobs like imageScaleFactor and outputStride let developers trade accuracy for frame rate directly in JavaScript. Google later superseded it with MediaPipe Pose/BlazePose, and the PoseNet packages are now unmaintained history — but they defined the browser pose API everything since has copied.

Specs

Context window

Image/video input (configurable resolution; 17 COCO keypoints)

Pricing

Free — Apache-2.0 model and code

Free quota

N/A (self-hosted)

API access

No hosted API — self-hosted TensorFlow.js/TF Lite weights

Weights

Open weights

Released

Not disclosed

Benchmarks

Provider-claimed figures belong to the underlying PersonLab model (arXiv 1803.08225): 0.665/0.687 COCO test-dev keypoint AP, single/multi-scale. PoseNet itself was never published with standalone benchmarks; a third-party mobile comparison (Jo & Kim, 2022) measured 97.6% accuracy in a small-sample setting — third-party, not a standard benchmark.

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

Use cases

browser-pose-estimationwebcam-pose-demoslegacy-pose-pipelineseducation

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics, metrics-v1). Evidence: official Google TF.js release with Apache-2.0 license; built on the peer-reviewed PersonLab paper (ECCV 2018) with published COCO numbers; widely deployed in its era. Strengths: genuinely open, historically important, simple API. Docked for: (1) retired by Google in favor of MediaPipe Pose — no maintenance; (2) benchmark figures belong to PersonLab, not PoseNet itself. License: Apache-2.0.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.0/10
Maintenance · recency, releases, issue load
3.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · Google TensorFlow.js pose model (2018), single + multi-person via PersonLab greedy decoding.
  • · Benchmark figures (0.665/0.687 COCO AP) belong to the underlying PersonLab model, not PoseNet itself.
  • · Retired by Google in favor of MediaPipe Pose; packages unmaintained but downloadable.

Sources