Yoga Pose
summary
Demo uses TensorFlow.js and React to run pose estimation entirely in the browser, using PoseNet (mobile net v1) to return 17 keypoints and draw overlays with the Canvas API. Pose scoring converts keypoints into XY vectors and compares them with cosine similarity against gold‑standard poses; a live high score of 38 is shown. ResNet50 proves too slow (≈10–15 fps), camera permissions and no tests cause reliability issues, but the browser pipeline otherwise works.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:12About Bryan Andy
- 3:36Why are you here
- 4:31React Hooks
- 5:46Tensorflow
- 7:02PoseNet
- 8:23Models
- 9:22Face Detection
- 9:37Pose Code
- 10:33Keypress
- 11:44Canvas
- 13:06Vectors
- 14:48Adding Vectors
- 15:45Mountain Pose Demo
- 16:45PoseNet Demo
- 17:21cosine similarity
- 17:47Andys challenge
- 22:09Design
- 23:53Potential Next Steps
related talks
05mIncredible things your community is creating with TensorFlow.js
30mPoseDance: Build a TikTok Trainer
30mBuilding Machine Learning Models on Edge with React
51mRealTime Exoskeleton-Mapping and Body Segmentation on the Browser with Tensorflow.js
54mDeep Learning in JS
15mMachine learning for next gen web apps with TensorFlow.js
24mTensorFlow.js + Node
22mStrike a Pose: Gesture Recognition in JS with Machine Learning