| Generatable or reverse engineerable personal data? |
No — SOMA generates anonymous 3D body meshes defined by continuous numerical shape coefficients. The canonical mesh topology is fixed and the released PCA shape space represents aggregate population-level statistics; it does not encode any specific individual's biometric identity and cannot be reverse-mapped to a participant. Shape coefficients do not correspond to real, identifiable individuals unless explicitly constructed to do so by the caller. The model does not process or output images, video, or biometric identifiers at inference time. |
| Personal data used to create this model? |
Partial — the SOMA native shape PCA was computed from two purchased 3D body scan datasets: (1) SizeUSA, consisting of whole-body scans of consenting human participants collected by TC², and dataset was processed to remove personally identifying markers before NVIDIA received them. (2) TripleGangers, containing body scans of 303 consenting individuals purchased from TripleGangers. GarmentMeasurement shape data was derived synthetically from their model and contains no personal data. Additionally, Bones RigPlay is a motion capture dataset of 350,000 animation sequences recorded from real human performers, used to train the optional pose-dependent corrective MLP. However, the data was retargeted to a fixed skeleton, removing any person-specific biometric signals. |
| Was consent obtained for any personal data used? |
Yes — SizeUSA participants consented to the collection and use of their body scan data for research and commercial purposes as part of the TC² data collection protocol. TripleGangers participants consented to scanning under agreements that explicitly permit use of the scans for AI model development. |
| Description of methods implemented in data acquisition or processing, if any, to address the prevalence of personal data in the training data: |
SizeUSA scans were delivered to NVIDIA as anonymous 3D body meshes without face geometry and texture maps. TripleGangers scans include full-body and face geometry; however, TripleGangers' participant agreements explicitly permit use of the scans for AI model development. Both datasets were processed without participant names, contact information, or other personal data. PCA fitting was performed on the mesh data, producing statistical shape vectors that cannot be reverse-mapped to individual participants. The resulting shape space (64 principal components) represents population-level body shape variation and not any specific individual's geometry. Bones RigPlay motion sequences are used only as skeletal animation data for training the optional corrective MLP; the retargeting process discards performer-specific kinematics and no visual appearance, face, texture, or identity information from any real performers is included. |
| How often is dataset reviewed? |
Dataset is initially reviewed upon addition, and subsequent reviews are conducted as needed or upon request for changes. |
| Is a mechanism in place to honor data subject right of access or deletion of personal data? |
Not Applicable — SOMA's release artifacts (model weights, PCA components) do not store raw scan data or any personal data. The PCA shape components are aggregate statistical transforms computed over all participants; no individual scan can be recovered from the released model. Data subject rights requests pertaining to the underlying SizeUSA dataset should be directed to TC² in accordance with their privacy policy. |
| If personal data was collected for the development of the model, was it collected directly by NVIDIA? |
No — body scan data was collected by TC² (SizeUSA) and TripleGangers (two third-party data providers) and purchased by NVIDIA under commercial licenses. NVIDIA did not directly collect body scans. |
| If personal data was collected for the development of the model by NVIDIA, do you maintain or have access to disclosures made to data subjects? |
Not Applicable — data was collected by TC² (SizeUSA) and TripleGangers; NVIDIA holds the commercial license agreements but not the participant consent forms, which remain with the respective data providers. |
| If personal data was collected for the development of this AI model, was it minimized to only what was required? |
Yes — only the 3D mesh geometry necessary for fitting the shape PCA were used. No facial texture, personal identifiers, or fine-grained biometric data were used. |
| Was data from user interactions with the AI model (e.g. user input and prompts) used to train the model? |
No |
| Is there provenance for all datasets used in training? |
Yes — SizeUSA: commercially licensed from TC²; TripleGangers: commercially licensed from TripleGangers (303 individuals); GarmentMeasurement: internally derived synthetic dataset using their source code which is released with GPL 3.0 license; Bones RigPlay: commercially licensed (purchased by NVIDIA), used for the optional pose corrective MLP. |
| Does data labeling (annotation, metadata) comply with privacy laws? |
Yes |
| Is data compliant with data subject requests for data correction or removal, if such a request was made? |
Not applicable with the released model — raw scan data is not included in the released artifacts. Requests relating to the source dataset should be directed to TC² or TripleGanger. |
| Applicable Privacy Policy |
https://www.nvidia.com/en-us/about-nvidia/privacy-policy/ |
