Learning Layered Implicit Model for 3D Avatar Clothing Representation

Keyu CHEN, Jingbo Wang, Wayne Wu, Bo Dai

Keywords: Geometric Representation, 3D Cloth, Human Body, Implicit Surface

TL;DR: We present a novel 3D cloth represention, i.e., a neural implicit surface model conditioned on volumetric SMPL prior, to capture realistic clothes from raw scans.

Abstract:

Modeling 3D clothed avatars is a popular topic in the computer graphics and vision area. Due to the complicated nature of realistic garments, the most concerned issue is how to represent 3D cloth shapes efficiently and effectively. A desirable cloth model is expected to preserve high-quality geometric details while establishing essential correspondences between clothes and animation-ready templates. However, by far there is no such a 3D cloth representation that can simultaneously satisfy these two requirements.
In this work, we thus formulate a novel 3D cloth representation that integrating the neural implicit surface with a statistical body prior.
Different from previous methods using explicit cloth primitives conditioned on the SMPL surface, we adopt a two-layer implicit function to capture the coarse and fine levels of cloth displacements, based on a parametric SMPL volume space. Our approach is aware of the underlying statistical minimal body shapes, and is also capable of modeling challenging clothes like skirts.
To evaluate the geometric modeling capacity of our 3D cloth representation, we conduct both qualitative and quantitative experiments on raw scans, which indicate superior performance over existing 3D cloth representations.
The effectiveness and flexibility of our 3D cloth representation is further validated in downstream applications, e.g. 3D virtual try-on.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)