Anny Generates Accurate 3D Body Models From Just 8 Questions, No Photos or GPU Required
Key Takeaways
- ▸A simple 8-question questionnaire can generate accurate 3D body models with 0.3 cm height accuracy and 3-4 cm circumference accuracy, rivaling photo-based methods
- ▸The physics-aware small MLP model runs efficiently on CPU in milliseconds, requiring no GPU and addressing privacy and user experience concerns
- ▸The breakthrough builds on research showing height + weight can predict body measurements well, but adds body-shape, build-type, and other parameters to handle body variation that simple regressions miss
Summary
Anny has developed a breakthrough questionnaire-based approach to generate detailed 3D body models using only eight questions, eliminating the need for photos or GPU processing. The method leverages a small multilayer perceptron trained with physics-aware loss functions, capable of estimating 58 body parameters with high accuracy: height within 0.3 cm, mass within 0.3 kg, and body measurements (bust/waist/hips) within 3-4 cm—matching or exceeding traditional photo-based pipelines. The approach was inspired by research showing that height and weight alone can predict body measurements surprisingly well, but Anny enhanced this foundation by incorporating additional signals like build type (athletic vs. soft), body shape distribution, and cup size to account for body variation at fixed height/weight combinations.
The questionnaire-based method addresses critical user experience and privacy concerns by eliminating the need for users to spend time finding properly-lit photos in appropriate clothing, while also reducing computational costs and speeding up the digital twin creation process. The approach runs efficiently on CPU in milliseconds, making it accessible and scalable. In developing the system, Anny discovered and corrected a mass calculation inconsistency in their existing model and refined their understanding of how muscle density affects body volume—insights that improve overall accuracy across their platform.
- The questionnaire approach eliminates friction from the digital twin creation process while improving measurement conventions to align with ISO 8559-1 anatomical standards
Editorial Opinion
This is an elegant technical solution that prioritizes user privacy and accessibility without sacrificing accuracy. By moving away from photo-based body reconstruction, Anny has created a faster, more inclusive system that respects user preferences while maintaining scientific rigor through physics-aware model training. The insight that simple anthropometric measurements combined with intuitive shape descriptors can outperform complex computer vision approaches challenges assumptions in the industry and suggests there's room for simpler, more practical solutions in 3D body modeling.
