DeepNAG: Deep Non-Adversarial Gesture Generation
summary
A generator-only training method replaces the discriminator by minimizing a differentiable combination of soft DTW and the average Hausdorff distance, producing DeepNAG synthetic gesture sequences. Evaluations augmenting training sets for 5 recognizers across 6 datasets show reduced recognition error often to perfect accuracy in some cases, and training runs up to 17× faster. A 180-participant HYPE∞ study finds DeepNAG samples judged more realistic than GAN outputs, achieving hyperrealism (>50) on the Leap Motion set.
AI-written summary · about
related talks
06mGAN-based Style Transformation to Improve Gesture-recognition Accuracy
19mA Rapid Prototyping Approach to Synthetic Data Generation For Improved 2D Gesture Recognition
43mHitchhiker’s Guide to Generative Adversarial Networks (GANs)
03mBuilding artificial people: Endless possibilities and the dark side
40mOverview of Generative Adversarial Networks (GANs)
37mHarnessing the power of Generative Adversarial Networks (GANS) for supervised learning
44mGenerative Adversarial Networks (GANs) and Stable Diffusion
13mNavigating Generative AI and Its Risks