3D Digital Avatars with Machine Learning
summary
Presents concrete systems: a pipeline that recovers a 3D garment and SMPL body parameters from a single phone image and runs about 10 frames per second, using image segmentation, normal prediction and a template-deformation step to reproduce fine wrinkles from 2D supervision; an animation model that learns per-vertex offsets in a two-stage network (coarse fit then wrinkle refinement) trained on physics‑based synthetic data; and a follow-up that learns garment deformation with data-free physics-based training, producing plausible wrinkles but requiring per-garment training and still missing soft‑tissue realism.
AI-written summary · about
chapters (7)
- 0:00Introduction to 3D avatars
- 3:08Modeling the human body
- 13:00Adding clothing to avatars
- 21:10Virtual try-on with AI
- 31:50Learning without data
- 37:30Human appearance synthesis
- 42:46Conclusion and Q&A
related talks
27mWorking Smart - Machine Learning in Art Asset Creation for the Games Industry
50mFuture of digital humans
12mOutline and Detail: A Semantic-Driven Framework for Layered 2D Character Generation
22mEverything a Swift Dev Needs to Know About Machine Learning
06mCoolMoves: User Motion Accentuation in Virtual Reality
46mGrowing Objects
22mPrompts to fashion: AI-powered 3D-printed garments
15mMeCap: Whole-Body Digitization for Low-Cost VR/AR Headsets