HIT: Estimating Internal Human Implicit Tissues from the Body Surface

Figure 1 . Left half: From volumetric human MRI scans, we learn to segment human internal tissues: subcutaneous adipose tissue (yellow), intra-muscular and visceral adipose tissue (blue), lean tissue (red), and long bones (white). We segment the MRI to extract a point cloud of the human body surface (red rings) to which we fit a human body model (SMPL, gray mesh). From this internal and external paired data, we learn Human Implicit Tissues (HIT), an implicit volumetric model that predicts the type and location of internal tissue. Right half: input body (blue mesh) and predicted tissues: subcutaneous adipose tissue (yellow) and lean tissue (red). We use OSSO [32] to infer the bones.