← ArchivePaper2021

SMPLicit: Topology-aware Generative Model for Clothed People

Enric Corona, Albert Pumarola, Guillem Alenya, Gerard Pons-Moll, Francesc Moreno-Noguer

An implicit function conditioned on SMPL parameters and a semantically interpretable latent code generates clothing of diverse topologies including open jackets, skirts, and shoes.

Open the paper ↗Trace in the lineage graph

Abstract

In this paper we introduce SMPLicit, a novel generative model to jointly represent body pose, shape and clothing geometry. In contrast to existing learning-based approaches that require training specific models for each type of garment, SMPLicit can represent in a unified manner different garment topologies (e.g. from sleeveless tops to hoodies and to open jackets), while controlling other properties like the garment size or tightness/looseness. We show our model to be applicable to a large variety of garments including T-shirts, hoodies, jackets, shorts, pants, skirts, shoes and even hair. The representation flexibility of SMPLicit builds upon an implicit model conditioned with the SMPL human body parameters and a learnable latent space which is semantically interpretable and aligned with the clothing attributes. The proposed model is fully differentiable, allowing for its use into larger end-to-end trainable systems. In the experimental section, we demonstrate SMPLicit can be readily used for fitting 3D scans and for 3D reconstruction in images of dressed people. In both cases we are able to go beyond state of the art, by retrieving complex garment geometries, handling situations with multiple clothing layers and providing a tool for easy outfit editing.

How to read this

Category: Method: a generative model for clothed human geometry
Contributions: A single generative model that represents many garment topologies (sleeveless tops, hoodies, open jackets, skirts, shoes, hair) in a unified way
An implicit model conditioned on SMPL body parameters plus a semantically interpretable, attribute-aligned latent space controlling size and tightness
A fully differentiable formulation usable for fitting 3D scans and reconstructing dressed people from images
Context: Builds on learning-based clothed-body modeling (Ma et al.'s CAPE, Learning to Dress 3D People in Generative Clothing) and extends the SMPL body model with an implicit, topology-aware clothing representation.Builds on: Learning to Dress 3D People in Generative Clothing
Correctness: Strength is topology flexibility from a single model and differentiability for downstream fitting; reported gains are demonstrated on 3D scan fitting and image-based reconstruction, so generalization beyond those settings and to extreme poses or unusual garments should be read cautiously.
Clarity: High-level idea (implicit clothing conditioned on SMPL plus interpretable latent) is accessible; the implicit-function training and conditioning details reward a second pass.
How to read it: First pass for the representation idea and what the latent code controls; second pass on the implicit formulation and fitting pipeline if you plan to reuse or extend it.

Builds on

Learning to Dress 3D People in Generative Clothing 2020

Built upon by

Nothing yet.

Related work

SCANimate: Weakly Supervised Learning of Skinned Clothed Avatar Networks 2021 / CVPR
N-Cloth: Predicting 3D Cloth Deformation with Mesh-Based Networks 2022 / CGF
SNUG: Self-Supervised Neural Dynamic Garments 2022 / CVPR
Motion Guided Deep Dynamic 3D Garments 2022 / SIGGRAPH Asia

Keywords

smplicit clothed people generative cloth model garment topology implicit function network unsigned distance field smpl latent space marching cubes linear blend skinning 3d reconstruction virtual try-on auto-decoder deepsdf body occlusion map pose-dependent deformation 3d scan fitting tailornet

This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →