PoseDance Build a TikTok Trainer
summary
A web app called PoseDance compares an exported TikTok MP4 to live webcam poses and scores how closely the user matches the moves. It runs PoseNet (17 keypoints) via TensorFlow.js with two PoseNet models (one for the video, one for the webcam), keeps frames for post-run scoring (scores are negative deltas, closer to zero is better), uses Azure Functions + PlayFab for auth and leaderboard, and is deployed to Azure Static Web Apps; models take ~7–10s to load.
AI-written summary · about
chapters (17)
- 0:00Introduction and welcome
- 2:24What is TikTok
- 3:49PoseNet basics
- 6:39PoseNet under the hood
- 11:09Building PoseDance
- 13:56Inclusive design considerations
- 14:17App architecture overview
- 17:28Interface and setup
- 19:11Asynchronous loading
- 20:34Pose detection frame
- 21:58Drawing skeletons
- 23:03Scoring mechanism
- 24:05Backend with PlayFab
- 25:28Leaderboard implementation
- 26:30Deployment with Azure
- 29:19Live demo of PoseDance
- 31:55Q&A session
related talks
23mCounting exercise repetition using live camera video
10mPoseTween: Pose-driven Tween Animation
16mVideo Personalization for Strength Training
21mPosePIE: Replace Your Keyboard and Mouse With AI-Driven Gesture Control
13mVisMimic: Integrating Motion Chain in Feedback Video Generation for Motor Coaching
26mSelf-Portraits in JavaScript
33mSee More, Be More
17mTransforming access to healthcare through Web AI