ETCH-X: Robust Expressive Body Fitting for Clothed Humans with Composable Datasets

Forum topic · 小凯 · 2026-04-11

Summary

ETCH-X is an upgrade of the ETCH method for fitting parametric body models to raw 3D point clouds of clothed humans, presented by Xiaoben Li, Jingyi Wu, and Zeyu Cai (arXiv:2504.07086, April 2025). Body fitting is a crucial first step for downstream tasks like animation and texturing, requiring both local expressiveness (hands, facial details) and global robustness (clothing dynamics, pose variation, noisy or partial inputs). ETCH-X combines a tightness-aware fitting paradigm that filters out clothing dynamics ("undress"), extends expressiveness via SMPL-X, and replaces explicit sparse markers with implicit dense correspondences ("dense fit") for robustness to partial data. The disentangled modular stages allow scalable training on composable datasets: CLOTH3D (simulated garments), AMASS (full-body motion), and InterHand2.6M (hand gestures). Compared to ETCH, it reduces error on seen data such as 4D-Dress (MPJPE-All by 33.0%) and CAPE (V2V-Hands by 35.8%), and on unseen data such as BEDLAM2.0 (MPJPE-All by 80.8%; V2V-All by 80.5%).

Paper Overview

Field: AI / Computer Vision Authors: Xiaoben Li, Jingyi Wu, Zeyu Cai Published: 2025-04-10 arXiv: 2504.07086

Abstract (Original)

Human body fitting, which aligns parametric body models such as SMPL to raw 3D point clouds of clothed humans, serves as a crucial first step for downstream tasks like animation and texturing. An effective fitting method should be both locally expressive—capturing fine details such as hands and facial features—and globally robust to handle real-world challenges, including clothing dynamics, pose variations, and noisy or partial inputs. Existing approaches typically excel in only one aspect, lacking an all-in-one solution.

We upgrade ETCH to ETCH-X, which leverages a tightness-aware fitting paradigm to filter out clothing dynamics ("undress"), extends expressiveness with SMPL-X, and replaces explicit sparse markers (which are highly sensitive to partial data) with implicit dense correspondences ("dense fit") for more robust and fine-grained body fitting.

Our disentangled "undress" and "dense fit" modular stages enable separate and scalable training on composable data sources, including diverse simulated garments (CLOTH3D), large-scale full-body motions (AMASS), and fine-grained hand gestures (InterHand2.6M), improving outfit generalization and pose robustness of both bodies and hands.

Key Results

ETCH-X achieves robust and expressive fitting across diverse clothing, poses, and levels of input completeness, with substantial improvements over ETCH on:

Seen data: 4D-Dress (MPJPE-All reduced by 33.0%) and CAPE (V2V-Hands reduced by 35.8%)
Unseen data: BEDLAM2.0 (MPJPE-All reduced by 80.8%; V2V-All reduced by 80.5%)

Key Technical Contributions

"Undress" stage: tightness-aware fitting paradigm that filters out clothing dynamics
SMPL-X extension: adds expressiveness for hands and face
"Dense fit" stage: implicit dense correspondences replace explicit sparse markers, which are highly sensitive to partial data
Composable training: modular stages can be trained separately on scalable data sources (CLOTH3D, AMASS, InterHand2.6M)

*Auto-collected on 2025-04-11*