Digital Humans, Avatars & Virtual Try-On
Benchmarking and Improving 3D Human Body Model Robustness Across Underrepresented Demographics and Non-Standard Morphologies
Generated automatically from the limitations stated in 4 papers (ECCV, CVPR, ICCV), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.
The problem
Existing 3D statistical body models and avatar pipelines are predominantly trained on narrow demographic subsets (such as adult European CAESAR scans) and evaluated solely on standard normative body templates. Consequently, downstream avatars and virtual try-on methods systematically produce distorted geometry, unnatural surface artifacts, and inaccurate anthropometric measurements on children, adolescents, diverse global populations, individuals with disabilities, and non-standard body proportions. Because existing benchmarks do not measure generalization across diverse body shapes, failure modes on these populations remain unquantified and unaddressed.
Why it matters
Enables avatar generation, digital human modeling, and virtual try-on systems to reliably represent global populations and non-standard body types without anatomical distortion or catastrophic measurement errors.
Ways to approach it
Prior-work checks are free with an account. Results someone already ran are shown to everyone.
- 1
Construct a standardized evaluation benchmark composed of diverse anthropometric scans spanning underrepresented demographics (children, teens, diverse global cohorts, and atypical body proportions) and measure the surface fitting error, measurement accuracy, and geometric distortion of standard parametric models (e.g., SMPL, SMPL-X, GHUM).
- 2
Develop a shape-prior regularization and anatomy-preserving deformation layer that constrains local limb and surface proportions against unnatural stretching or elongation when fitting out-of-distribution body morphologies from images or scans, evaluating chamfer distance and anatomical plausibility metrics.
- 3
Build a multi-scale compositional shape space that decouples skeletal bone proportions from soft-tissue demographic variations to enable explicit synthesis and fitting of extreme and underrepresented body types without requiring massive new dense 3D scan corpora.
Have a different approach?
Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.
Why it might fail
The project could fail if acquiring diverse 3D ground-truth scan data of protected or underrepresented demographics (e.g., minors or clinical populations) is blocked by institutional review or licensing constraints, forcing reliance on synthetic approximations that fail to reflect real anatomical variance.
Evidence
Each paper's own statement of the limitation, verbatim.
- Design2GarmentCode: Turning Design Concepts to Tangible Garments Through Program SynthesisCVPR 2025
Evaluated with a single standard body model, so made-to-measure / body-shape generalization is untested in the reported metrics
- ATLAS: Decoupling Skeletal and Shape Parameters for Expressive Parametric Human ModelingICCV 2025
15,000 training subjects do not span the full range of human shape variation, so coverage of the population remains incomplete
- GarmentCodeData: A Dataset of 3D Made-to-Measure Garments With Sewing PatternsECCV 2024
Body measurements can be inaccurate on extreme body shapes due to surface fit errors; statistical body model is biased toward healthy European/North American adults (only the CAESAR European subset), excluding geographic variation, disabilities, teens, and children
- Combining Generative and Geometry Priors for Wide-Angle Portrait CorrectionECCV 2024
Corrected body regions can be unnatural (e.g., visibly elongated feet in qualitative results), as no body prior is modeled
Nearest existing work
- ATLAS: Decoupling Skeletal and Shape Parameters for Expressive Parametric Human ModelingICCV 2025
- Towards Explicit Exoskeleton for the Reconstruction of Complicated 3D Human AvatarsICCV 2025
- A Focused Human Body Model for Accurate Anthropometric Measurements ExtractionCVPR 2025
- PHD: Personalized 3D Human Body Fitting with Point DiffusionICCV 2025
- BLSM: A Bone-Level Skinned Model of the Human MeshECCV 2020
- CLOTH3D: Clothed 3D HumansECCV 2020
- SCANimate: Weakly Supervised Learning of Skinned Clothed Avatar NetworksCVPR 2021
- Joint Optimization for Multi-Person Shape Models from Markerless 3D-ScansECCV 2020
- SUPR: A Sparse Unified Part-Based Human RepresentationECCV 2022
- AGORA: Avatars in Geography Optimized for Regression AnalysisCVPR 2021
- Cloth2Body: Generating 3D Human Body Mesh from 2D ClothingICCV 2023
- MeasureXpert: Automatic Anthropometric Measurement Extraction from Two Unregistered, Partial, Posed, and Dressed Body ScansICCV 2025
- X-Avatar: Expressive Human AvatarsCVPR 2023
- OSSO: Obtaining Skeletal Shape From OutsideCVPR 2022
- BodyNet: Volumetric Inference of 3D Human Body ShapesECCV 2018
Related open problems
Digital Humans, Avatars & Virtual Try-On
Overcoming Parametric 3DMM Expressiveness and Mouth Interior Bottlenecks in Monocular Facial Avatars
Neural head avatar pipelines universally rely on parametric mesh models like FLAME as geometric proxies, inheriting their topological limitations. Because FLAME lacks internal oral anatomy (teeth and tongue) and fine-grained geometric detail, avatars suffer from severe projection artifacts such as teeth textures baked directly onto lip surfaces and distorted speech articulation. Furthermore, low-dimensional linear expression spaces cap the capture of dynamic micro-expressions, dynamic wrinkles, and non-parametric regions like hair.
Digital Humans, Avatars & Virtual Try-On
Joint Pose Refinement and Error-Tolerant Conditioning for Downstream Human Avatar Synthesis
Current avatar creation and virtual try-on pipelines treat SMPL/SMPL-X estimations, camera calibrations, and segmentation masks as fixed, ground-truth preconditions. When off-the-shelf estimators inevitably produce misaligned or out-of-distribution body parameters on in-the-wild images, these errors cascade into severe geometric distortion, stitching artifacts, and generation failure. Because existing downstream pipelines do not jointly optimize human body parameters or account for parameter uncertainty, high-fidelity human generation remains restricted to controlled studio captures with pristine initializations.
Digital Humans, Avatars & Virtual Try-On
Reducing Downstream Dependency of Garment/Avatar Generation on Upstream 2D and 3D Human-Parsing Predictions
Every pipeline in this space chains off-the-shelf estimators — pose keypoint detectors, DensePose, SMPL/SMPL-X fitting, segmentation/parsing models, SAM — and treats their outputs as ground truth. When these upstream predictions are noisy, fail on exaggerated poses, unusual garments, or garment-only imagery, the downstream model inherits or amplifies the errors (garment leaking, misinitialized garment geometry, failed reposing). No one measures or controls how error propagates through this preprocessing chain, so each new pipeline is brittle in its own undocumented way.