ETCH-X: Robust and Expressive Body Fitting for Clothed Humans

Forum topic · 小凯 · 2026-04-12

Summary

ETCH-X (arXiv:2504.07950) is a computer vision method that aligns expressive parametric body models such as SMPL-X with raw 3D point clouds of clothed humans. It upgrades the earlier ETCH method with three ideas: a tightness-aware fitting paradigm that filters out clothing dynamics ('undress'), extension to SMPL-X for finer hand and facial detail, and replacement of explicit sparse markers—which are sensitive to partial inputs—with implicit dense correspondences ('dense fit'). The modular design allows scalable training on composable data sources, including simulated garments (CLOTH3D), large-scale body motion (AMASS), and fine-grained hand gestures (InterHand2.6M). Experimental results show significant improvements over ETCH: on seen datasets such as 4D-Dress (MPJPE-All +33.0%) and CAPE (V2V-Hands +35.8%), and on the unseen BEDLAM2.0 dataset (MPJPE-All +80.8%, V2V-All +80.5%), across varied clothing, poses, and input completeness levels.

ETCH-X: Robustify Expressive Body Fitting to Clothed Humans

Field: Computer Vision Authors: Xiaoben Li, Jingyi Wu, Zeyu Cai arXiv: 2504.07950

Abstract

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"), enabling more robust and fine-grained human body fitting.

Our decoupled "undress" and "dense fit" modular stages support scalable training on composable data sources, including diverse simulated garments (CLOTH3D), large-scale full-body motion (AMASS), and fine-grained hand gestures (InterHand2.6M), improving garment generalization and pose robustness for both body and hands.

Results

ETCH-X achieves robust and expressive fitting under diverse clothing, poses, and input completeness levels, with significant improvements over ETCH:

Seen datasets:
4D-Dress: MPJPE-All improved by 33.0%
CAPE: V2V-Hands improved by 35.8%
Unseen dataset:
BEDLAM2.0: MPJPE-All improved by 80.8%, V2V-All improved by 80.5%