BlazePose: On-device Real-time Body Pose tracking

We present BlazePose, a lightweight convolutional neural network architecture for human pose estimation that is tailored for real-time inference on mobile devices. During inference, the network produces 33 body keypoints for a single person and runs at over 30 frames per second on a Pixel 2 phone. This makes it particularly suited to real-time use cases like fitness tracking and sign language recognition. Our main contributions include a novel body pose tracking solution and a lightweight body pose estimation neural network that uses both heatmaps and regression to keypoint coordinates.

PDF Abstract

Datasets


Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
3D Pose Estimation Google-AR BlazePose Lite PCK@0.2 79.6 # 3
3D Pose Estimation Google-AR BlazePose Full PCK@0.2 84.1 # 2
3D Pose Estimation Google-AR OpenPose (body only) PCK@0.2 87.8 # 1
3D Pose Estimation Google-Yoga BlazePose Full PCK@0.2 84.5 # 1
3D Pose Estimation Google-Yoga BlazePose Lite PCK@0.2 77.6 # 3
3D Pose Estimation Google-Yoga OpenPose (body only) PCK@0.2 83.4 # 2

Methods


No methods listed for this paper. Add relevant methods here