HIT: Estimating Internal Human Implicit Tissues from the Body Surface

How to read this

Category

Method + dataset: implicit internal-anatomy model

Contributions

• HIT, an implicit volumetric function that classifies body-interior points as subcutaneous adipose tissue, lean tissue, or long bone given an external body shape.
• A learned volumetric deformation field correcting soft-tissue changes between upright and supine MRI positions.
• Parameterization by SMPL so internal structures deform plausibly when the body is reposed or reshaped; dataset (260 female, 182 male full-body MRI) and model released publicly.

Context

Continues the authors' line on inferring internal structure from the body surface (OSSO, SKEL) and is built on the SMPL body model (Loper et al., 'SMPL: A Skinned Multi-Person Linear Model').Builds on: OSSO: Obtaining Skeletal Shape from Outside · From Skin to Skeleton: Towards Biomechanically Accurate 3D Digital Humans · SMPL: A Skinned Multi-Person Linear Model

Correctness

Trained and grounded on full-body MRI scans with a stated subject count, so a reader should keep in mind the demographic and posture range of that scan set and that tissue inference is a learned classification, not a per-subject medical measurement.

Clarity

Accessible given familiarity with SMPL and implicit functions; a first pass conveys the inputs, tissue classes, and SMPL coupling, with a second pass for the deformation field.

How to read it

First pass for what HIT infers and how SMPL parameterization drives reposing; second pass on the upright-to-supine deformation field and training data if you plan to use the released model.