A Rapid Prototyping Approach to Synthetic Data Generation For Improved 2D Gesture Recognition
summary
Introduces gesture path stochastic resampling, a two-step rapid-prototyping generator that randomly resamples points along a stroke then renormalizes inter-point distance to lengthen or shorten subpaths, implemented in plain code and running at about 6–8 microseconds per sample on a Surface Pro. Trains $1-family recognizers with 64 synthetic samples per real and reports typical error reductions of 30–80% (examples: Perlin 42%, Sigma log‑normal 61%, GPSR ≈80%), evaluated across public gesture datasets and mixed‑writer protocols.
AI-written summary · about
chapters (22)
- 0:00Intro
- 0:26What are Gestures
- 1:55Better Gesture Recognition
- 2:09Synthetic Data
- 2:35Presentation Outline
- 3:09Synthetic Variation
- 4:11Noise Models
- 5:00Domain Knowledge
- 5:20Visualization
- 5:45Difficulties
- 6:09Rapid Prototyping
- 7:22Example
- 8:44Resampling
- 10:13Synthetic Distribution
- 11:01Synthetic Features
- 11:53Synthetic Distributions
- 12:04Results
- 12:20Protocol
- 12:52Points of Interest
- 13:20Testing Results
- 14:48Conclusion
- 16:09Questions
related talks
10mDeepNAG: Deep Non-Adversarial Gesture Generation
10mA large, crowdsourced evaluation of gesture generation systems on common data: The GENEA ...
06mGAN-based Style Transformation to Improve Gesture-recognition Accuracy
06mPantomime: Mid-Air Gesture Recognition with Sparse Millimeter-Wave Radar Point Clouds
22mStrike a Pose: Gesture Recognition in JS with Machine Learning
28mStrike a Pose - Gesture recognition in JavaScript w/ ML & Arduino
06mCrossGR: Accurate and Low-cost Cross-target Gesture Recognition Using Wi-Fi
19mBackHand: Sensing Hand Gestures via Back of the Hand