From measurements to shape parameters

[arianp95]

Is it possible to obtain the shape parameters from the body measurements?

[DavidBoja]

Hi,

sure.
You can randomly sample lots of SMPL shapes, get the body models with those shapes, and measure them using this codebase. Then you have your inputs (measurements) and your outputs (shape parameters) and you can learn a model that maps one to the other. I believe the model could be fairly simple, like linear regression. It should be fairly straightforward.

For a better model, you could use the CAESAR dataset and fit the SMPLs to the 3D scans and use the provided manual measurements.

There are works that have done this so you can check them out:

• Home 3D body scans from noisy image and range data
• The Virtual Caliper: Rapid Creation of Metrically Accurate Avatars from 3D Measurements
• A statistical model of human pose and body shape