ReMu
Reconstructing Multi-layer 3D Clothed Human from Images

BMVC 2025 Paper

Onat Vuran Hsuan-I Ho

TL;DR

Project Teaser Image

ReMu reconstructs multi-layer garments from Image Layers. (a) Given a set of images captured by a static RGB camera, (b) our method reconstructs 3D garments in layers that fit on a unified human body. These layers are optimized to avoid possible inter-penetrations, (c) making the resulting 3D garments suitable for various downstream applications, such as clothing simulation.

Abstract

The reconstruction of multi-layer 3D garments typically requires expensive multi-view capture setups and specialized 3D editing efforts. To support the creation of life-like clothed human avatars, we introduce ReMu for reconstructing multi-layer clothed humans in a new setup, Image Layers, which captures a subject wearing different layers of clothing with a single RGB camera. To reconstruct physically plausible multi-layer 3D garments, a unified 3D representation is necessary to model these garments in a layered manner. Thus, we first reconstruct and register each garment layer in a shared coordinate system defined by the canonical body pose. Afterwards, we introduce a collision-aware optimization process to address interpenetration and further refine the garment boundaries leveraging implicit neural fields. It is worth noting that our method is template-free and category-agnostic, which enables the reconstruction of 3D garments in diverse clothing styles. Through our experiments, we show that our method achieves competitive performance compared to category-specific methods and reconstructs penetration-free 3D clothed humans.

Method

Method Architecture Diagram

Given a set of k-layered Image Layers Ik, we register SMPL-X body models B and reconstruct watertight meshes M from images. We then segment out 3D garments S through multi-view parsing. Next, we deform 3D meshes to the canonical body pose Bc through inverse LBS. These aligned garments are optimized to remove inter-layer penetrations as M'. Finally, we fit implicit neural fields f to refine the garment surface geometry and boundaries.

    Technical Contributions:

  • Garment Registration: The first key aspect of our method.
  • Penetration Removal: The second innovative feature.
  • Garment Refinement: How we handle a specific challenge.

Results on 4D-DRESS

Input Images

Input Image

SMPLicit

ClothWild

ISP

Ours

Input Images

Input Image

SMPLicit

ClothWild

ISP

Ours

In-the-wild Capture

In-the-wild scene 1
In-the-wild scene 2
In-the-wild scene 3
In-the-wild scene 4

References

  • Ho et. al, "SiTH: Single-view Textured Human Reconstruction with Image-Conditioned Diffusion." CVPR, 2024.
  • Kim et. al, "GALA: Generating Animatable Layered Assets from a Single Scan." CVPR, 2024.
  • Wang et. al, "4D-DRESS: A 4D Dataset of Real-world Human Clothing with Semantic Annotations." CVPR, 2024.
  • Corona et. al, "SMPLicit: Topology-aware Generative Model for Clothed People." CVPR, 2021.
  • Moon et. al, "ClothWild: 3D Clothed Human Reconstruction in the Wilde." ECCV, 2022.
  • Li et. al, "ISP: Multi-Layered Garment Draping with Implicit Sewing Patterns." NeurIPS, 2023.
  • Citation

    @inproceedings{vuran2025remu, title = {ReMu: Reconstructing Multi-layer 3D Clothed Human from Images}, author = {Vuran, Onat and Ho, Hsuan-I}, booktitle = {British Machine Vision Conference (BMVC)}, year = {2025} }