```html

Body Modeling Under Clothing: New ETCH Method Significantly Improves Accuracy

Accurately fitting a body model to 3D point clouds of clothed humans is a fundamental challenge in fields like virtual try-on, animation, and medical imaging. Traditional optimization-based methods use multi-stage pipelines that are sensitive to pose initialization. Learning-based methods, on the other hand, often struggle with generalization across different poses and clothing styles. A new approach called ETCH (Equivariant Tightness Fitting for Clothed Humans) promises significant improvements in this area.

ETCH utilizes a novel method that estimates the mapping from clothing surface to body surface through locally approximated SE(3)-equivariance. It encodes clothing tightness as a displacement vector from the clothing surface to the underlying body. After this mapping, pose-invariant body features are used to determine sparse body markers. This simplifies the fitting of the clothed human to a fitting task of inner body markers.

Equivariance plays a crucial role because it ensures the consistency of calculations across different rotations and translations in space. This allows for a more robust estimation of body shape, regardless of the person's pose or the type of clothing.

To evaluate ETCH's performance, extensive experiments were conducted on established datasets such as CAPE and 4D-Dress. The results show that ETCH significantly outperforms existing methods, both those that consider clothing tightness and those that do not, in terms of body fitting accuracy with loose clothing (16.7% ~ 69.5%) and shape accuracy (49.9% on average).

Particularly noteworthy is ETCH's ability to reduce directional errors in one-shot scenarios (or out-of-distribution settings) by (67.2% ~ 89.8%). This indicates strong generalization capabilities, which also extend to challenging poses, unknown body shapes, loose clothing, and non-rigid dynamics.

The developers of ETCH plan to release the code and models soon for research purposes. This opens up the opportunity for the research community to further investigate the method and adapt it for various applications. The promising results of ETCH suggest that this approach can make a significant contribution to improving body modeling under clothing and has the potential to advance the development of more realistic virtual avatars and more accurate medical imaging techniques.

The improved accuracy and generalization ability of ETCH compared to previous methods opens up new possibilities in various application areas. These include:

- Virtual Try-on: More realistic simulation of clothing fit on different body types. - Animation: Creation of more believable movements of clothed figures in movies, games, and virtual environments. - Medical Imaging: Improved analysis of body shapes and movements under clothing for diagnostic purposes. Bibliography: https://arxiv.org/abs/2503.10624 https://arxiv.org/html/2503.10624 https://boqian-li.github.io/ETCH/ http://paperreading.club/page?id=291746 https://github.com/boqian-li/ETCH https://papers.cool/arxiv/cs.GR https://github.com/weihaox/awesome-digital-human https://boqian-li.github.io/ https://chatpaper.com/chatpaper/zh-CN/paper/120358 https://cofeed.app/categories/cs.GR ```