Roboflow

Best Pose Estimation Models

Compare the best 3 pose estimation models. 3 are open-weight, so you can self-host them for free under their licenses.

3 models · 3 open-weight · prices synced Aug 7, 2026

Open-Source Pose Estimation Models

3 models with downloadable weights you can self-host under their licenses (AGPL 3.0 and Apache 2.0).

YOLOv12
YOLOv12 is an attention-centric real-time object detection model developed by researchers at Tsinghua University, with the arXiv paper published in February 2025 under the AGPL-3.0 license. It introduces an Area Attention module that partitions feature maps into regions and applies self-attention within each region, reducing the quadratic complexity of full self-attention while capturing long-range dependencies. It also incorporates R-ELAN for improved feature aggregation and scaled residual connections for training stability.YOLOv12-L achieves 54.0% AP on COCO, while the YOLOv12-N variant achieves 40.5% mAP at 1.62ms latency on an NVIDIA T4 GPU. The model is built on the Ultralytics codebase, supporting detection, segmentation, and other standard YOLO tasks at competitive real-time speeds.
YOLOv8 Pose Estimation
YOLOv8 Pose Estimation is the keypoint detection variant of the YOLOv8 model developed by Ultralytics, released in April 2023 under the AGPL-3.0 license. It extends the YOLOv8 detection head to predict keypoint locations and visibility scores alongside bounding boxes, using a decoupled head for joint localization and keypoint regression. By default it targets the 17-keypoint COCO human pose skeleton, but can be configured for custom keypoint sets.YOLOv8 Pose shares the same architecture and size variants as the base detection model and achieves competitive performance on the COCO keypoints benchmark at real-time inference speeds. The model is deployable through Roboflow Inference and is suited for applications including sports analytics, ergonomics monitoring, gesture recognition, and human activity detection.
Google
MediaPipe
MediaPipe is an open-source framework developed by Google for building real-time machine learning pipelines across mobile, web, desktop, and edge platforms. First released in 2019, the framework uses a graph-based architecture where pre-built components called Calculators process streaming data such as images, video, and audio through configurable computation graphs. This design allows developers to compose perception pipelines from reusable building blocks without writing custom glue code between models. The current MediaPipe Tasks API replaces the earlier Solutions API and provides a unified cross-platform interface for vision, text, and audio.Rather than providing a single model, MediaPipe ships a suite of ready-to-use Tasks that wrap trained models for specific problems. These include MediaPipe Pose Landmarker for 33-point body landmark detection, Hand Landmarker for 21-point hand tracking, Face Landmarker which extends the earlier 468-point Face Mesh with blendshape outputs for facial expression, Selfie Segmentation for person-background separation, and Holistic Landmarker for combined body, hand, and face tracking. The Tasks prioritize on-device inference with low latency and support GPU acceleration where available, making the framework a common choice for mobile augmented reality, fitness and wellness applications, gesture-based interfaces, and accessibility features such as sign language recognition.

How to Choose a Pose Estimation Model

Pose models locate body joints and connect them into skeletons. Choosing one comes down to scene complexity, real-time requirements, and whether you need 2D or 3D output.

Toolkits like MediaPipe run pose at video rate on modest hardware, including phones, which is what fitness form feedback, gesture interfaces, and interactive installations need. They are strongest on one or a few clearly visible people; heavy occlusion and crowds are where they break down.

Crowded scenes force the top-down versus bottom-up choice: top-down methods (detect people, then find joints per person) are more accurate but scale in cost with the number of people; bottom-up methods find all joints at once and group them, holding speed steady in crowds. Frameworks like Detectron2 include keypoint heads for this class of problem.

Most models output 2D joints in image space; biomechanics, ergonomics, and motion capture need 3D lifting, which adds error and usually calibration. Decide based on the downstream measurement: counting reps needs 2D, measuring joint angles credibly usually needs 3D or multiple views.

The bottom line: Edge and interactive applications point to MediaPipe-class real-time models; crowded scenes and accuracy point to heavier top-down pipelines; let the downstream measurement decide 2D versus 3D.

Frequently Asked Questions About Pose Estimation Models

Pose estimation is the task of locating body joints such as shoulders, elbows, hips, and knees in images or video and connecting them into a skeleton that captures how a person or animal is positioned. Top-down methods detect people first and then find the joints of each person, while bottom-up methods find all joints and group them; keypoints are usually predicted as heatmaps and can be lifted from 2D to 3D. Accuracy is scored with keypoint metrics such as OKS (object keypoint similarity) or PCK. Pose estimation powers fitness form feedback, sports and biomechanics analytics, ergonomic assessment, fall detection, and markerless motion capture. This page lists 3 pose estimation models, including 3 open-weight options you can self-host.

It depends on your task and constraints. For fixed categories in production, a model fine-tuned on your own data typically beats any general-purpose model. Compare the pose estimation models on this page and try them on your own image to see which fits.

Yes. 3 of the 3 pose estimation models here are open-weight (for example YOLOv12, YOLOv8 Pose Estimation, and MediaPipe), free to self-host under their licenses (AGPL 3.0 and Apache 2.0).

This page lists all 3 pose estimation models in the Roboflow Playground catalog, all of them open-weight and free to self-host. Compare licenses, parameters, and release dates side by side, or open any model page for full details.