Research map / 3D vision / Face and shape
HIT: Estimating Internal Human Implicit Tissues from the Body Surface
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Builds a segmented full-body MRI dataset (442 scans) and trains HIT, an implicit SMPL-parameterized network that predicts internal tissue class (bone, lean, subcutaneous fat, empty) from body shape and pose.
Problem
Predicting the 3D location of internal human tissues (subcutaneous fat, lean tissue, long bones) from the external body surface, without MRI/CT at inference time.
Approach
Formulate internal anatomy inference as multi-tissue point classification in a SMPL canonical space, with learned deshaping, unposing (root-finding initialized by skinning weights) and volumetric decompression modules that undo the soft-tissue deformation from lying prone in an MRI scanner.
Key results
- HIT achieves Dice of 77.8 (LT), 57.7 (SAT), 45.5 (BT) for females and 81.0, 54.7, 52.2 for males, far above the Chance baseline (e.g. 51.4, 40.2, 3.9 female).
- Ablating the compression module (HITncmp) leaves compression in predictions for standing bodies, showing the decompression module is needed to generalize to new poses and shapes.
- Predicted tissue volumes are close to the average tissue quantity: relative volume error for HIT is roughly on par with the Chance baseline.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use HIT instead of the Chance baseline | HIT · Female test, LT | Dice score: 51.4 -> 77.8 | Helped |
| Use HIT instead of the Chance baseline | HIT · Female test, SAT | Dice score: 40.2 -> 57.7 | Helped |
| Use HIT instead of the Chance baseline | HIT · Female test, BT | Dice score: 3.9 -> 45.5 | Helped |
| Use HIT instead of the Chance baseline | HIT · Male test, LT | Dice score: 60.3 -> 81.0 | Helped |
| Use HIT instead of the Chance baseline | HIT · Male test, SAT | Dice score: 31.8 -> 54.7 | Helped |
| Use HIT instead of the Chance baseline | HIT · Male test, BT | Dice score: 4.1 -> 52.2 | Helped |
| Add the compression module (ablation) | HIT · Generalization to standing bodies | Qualitative compression plausibility | Helped |
| Use HIT instead of the Chance baseline | HIT · Female test, SAT | Relative tissue quantity difference (∆%): 6.7 -> 9.4 | Hurt |
Surprising findings
- Relative tissue volume error of HIT is about the same as a chance baseline predicting mean tissue fractions, suggesting it predicts average quantities rather than individual volumes.
Limitations
- Cannot validate the uncompression behavior inside the body without specific data or simulation.
- Precise tissue percentage prediction remains limited; tissue volumes are conditioned only on 10 SMPL shape parameters.
- IMVAT location is not modeled, and there is no explicit spatial-consistency loss.
Experimental setup
Datasets: HIT dataset (segmented full-body MRI scans with SMPL fits)
Benchmarks: HIT dataset test split
Metrics: Dice score relative volume error (delta %)
Baselines: Chance baseline HITncmp (no compression module) OSSO
Builds on: SMPL nnU-Net gDNA OSSO
Closest papers by meaning
- OSSO: Obtaining Skeletal Shape From Outside (CVPR 2022 · similarity 0.86)
- MosaicMRI: A Diverse Dataset and Benchmark for Raw Musculoskeletal MRI (NeurIPS 2026 · similarity 0.84)
- Point2SSM: Learning Morphological Variations of Anatomies from Point Clouds (ICLR 2024 · similarity 0.83)
- Anatomically Constrained Implicit Face Models (CVPR 2024 · similarity 0.82)
- Towards a Comprehensive, Efficient and Promptable Anatomic Structure Segmentation Model Using 3D Whole-Body CT Scans (AAAI 2025 · similarity 0.82)
- BodyMAP - Jointly Predicting Body Mesh and 3D Applied Pressure Map for People in Bed (CVPR 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.