Show HN: A 3D Body Scan for Nine Cents – Without SMPL

8 hours ago

SMPL is widely used in computer vision for human body reconstruction but has a non-commercial license, making it expensive and inaccessible for small startups.
The authors built a commercial pipeline using Naver's Anny and Meta's MHR, both with permissive licenses, to create a consumer-grade 3D body scanning solution.
Anny offers semantic shape parameters (e.g., gender, age, weight) and local blendshapes, while MHR provides high-quality meshes but with abstract coefficients.
The pipeline includes two input paths: a photo-based path using SAM 3D Body and MHR to Anny conversion, and a questionnaire path predicting Anny params without a GPU.
Measurement tuning is a key step to refine body models, aiming for less than 1 cm error by adjusting parameters based on known measurements.
Costs are low: about $0.09 per photo-based scan (optimizable to $0.03) and under $0.01 for the questionnaire path, with physics draping adding minimal expense.
Evaluation shows bust-waist-hips measurement errors of 5-8 cm initially, but tuning can improve accuracy, with challenges in absolute measurement recovery from photos.
The authors emphasize the importance of a measurement infrastructure and open-sourced tools, while noting UX hurdles in both photo and questionnaire methods.
Future improvements are planned, and the pipeline demonstrates a viable, low-cost alternative to SMPL for commercial use.

Hasty Briefsbeta