PoseDance: Build a TikTok Trainer
summary
Builds a web app that compares a TikTok video to webcam skeletons using PoseNet via TensorFlow.js, draws keypoints on canvas, and runs two simultaneous models (one for video, one for webcam) that take about ten seconds to load. Authentication and score submission use Azure Functions + PlayFab. Scoring is a simple per-keypoint subtraction producing mostly negative scores (top anonymous score ≈ 6); demo works but lags and scoring needs rework.
AI-written summary · about
chapters (24)
- 0:00Introduction
- 1:15What is TikTok
- 3:35Example Output
- 4:05Person Lab
- 5:24Behind the Scenes
- 8:06About PoseDance
- 9:15The Idea
- 10:02Ableism
- 11:26Building PoseDance
- 12:26Dependencies
- 13:12Embedding
- 14:01PoseNet
- 15:19Posing
- 16:04Drawing
- 16:23Scoring
- 17:24Backend
- 18:00Leaderboard
- 18:28Deployment
- 19:17Demo
- 21:28QA
- 22:57Life Problems
- 24:23Body Types
- 25:08Other Applications
- 26:52Rework Scoring
related talks
26mYoga Pose
05mIncredible things your community is creating with TensorFlow.js
51mRealTime Exoskeleton-Mapping and Body Segmentation on the Browser with Tensorflow.js
30mBuilding Machine Learning Models on Edge with React
54mDeep Learning in JS
23mCounting exercise repetition using live camera video
15mMachine learning for next gen web apps with TensorFlow.js
22mVue + TensorFlow.js: Building Your First ML Powered App