Research map / 3D vision / Human pose estimation
A Focused Human Body Model for Accurate Anthropometric Measurements Extraction
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.
Proposes a focused human body model that adds a trainable CNN/ResNet bypass network and a dynamic loss to a frozen SMPLer-X backbone to extract anthropometric measurements from images and LiDAR point clouds, and releases the Fashion-body dataset.
Problem
Accurate extraction of 3D anthropometric body measurements (chest, waist, hip, wrist, shoulder width) from images or point clouds is sensitive to camera angle, distance, posture, and clothing, and lacks paired multimodal benchmark data.
Approach
A lightweight SIREN-activated CNN/ResNet bypass network attached to a frozen SMPLer-X ViT-H backbone, trained with a dynamic loss that re-weights measurement-region versus non-measurement vertices, followed by ISO 8559-based measurement computation on the SMPL mesh.
Key results
- Our-BMViT reaches chest MAE 3.07 cm, waist 3.91 cm, hip 6.95 cm, wrist 0.41 cm on Fashion-body, beating HMR-BMViT and 4D-BMViT.
- On LiDAR data, Our-BMViT achieves chest MAE 3.07 cm vs 5.03 cm for Point2Mesh.
- Against commercial apps (3DLook, Verifyt) on iPhone 12 captures, Our-BMViT has chest MAE 3.32 cm vs 5.21 cm (3DLook) and wrist MAE 0.11 cm vs 0.33 cm (Verifyt).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add the Bypass Network and LDL to SMPLer-X (ablation) | SMPLer-X · Fashion-body | MAE across chest, waist, hip, wrist, and shoulder width (cm): Chest: 6.34 -> 3.07; waist: 9.84 -> 3.91; hip: 11.05 -> 6.95; wrist: 1.13 -> 0.41; shoulder width: 4.57 -> 2.03 | Helped |
| Add LDL (ablation) | Focused human body model with Bypass Network · Fashion-body | MAE across chest, waist, hip, wrist, and shoulder width (cm): Chest: 5.61 -> 3.07; waist: 9.22 -> 3.91; hip: 10.53 -> 6.95; wrist: 1.02 -> 0.41; shoulder width: 4.42 -> 2.03 | Helped |
| Add the Bypass Network (ablation) | Focused human body model with LDL · Fashion-body | MAE across chest, waist, hip, wrist, and shoulder width (cm): Chest: 4.73 -> 3.07; waist: 7.04 -> 3.91; hip: 10.36 -> 6.95; wrist: 1.00 -> 0.41; shoulder width: 4.48 -> 2.03 | Helped |
Surprising findings
- Verifyt outperforms the proposed method on hip measurements in the mobile-app comparison, while showing large waist error.
Limitations
- Fashion-body is small (30 paired subjects), and the mobile-app comparison uses only six subjects.
- Measurement is restricted to five measurements (chest, waist, hip, wrist, shoulder width).
- Reported dataset release is stated as a contribution, but no explicit link is given in the text.
Experimental setup
Datasets: Body Measurements Datasets Fashion-body
Benchmarks: Body Measurements Datasets Fashion-body LiDAR data from Mojave sensor
Metrics: MAE RPE AMAD
Baselines: Linear Regression NeuralAnthro HMR-BMViT 4D-BMViT Point2Mesh 3DLook YourFit Verifyt SMPLer-X
Builds on: SMPLer-X SMPL-X HMR SIREN
Closest papers by meaning
- ShapeBoost: Boosting Human Shape Estimation with Part-Based Parameterization and Clothing-Preserving Augmentation (AAAI 2024 · similarity 0.80)
- UMotion: Uncertainty-driven Human Motion Estimation from Inertial and Ultra-wideband Units (CVPR 2025 · similarity 0.80)
- Towards Fine-Grained HBOE with Rendered Orientation Set and Laplace Smoothing (AAAI 2024 · similarity 0.78)
- MUC: Mixture of Uncalibrated Cameras for Robust 3D Human Body Reconstruction (AAAI 2025 · similarity 0.78)
- Scaling Wearable Foundation Models (ICLR 2025 · similarity 0.78)
- AiOS: All-in-One-Stage Expressive Human Pose and Shape Estimation (CVPR 2024 · similarity 0.78)
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.