A. Polanco et al., "Skinning Weights Optimization through Data-Driven Approaches: Determining the Training Dataset", Proc. of 3DBODY.TECH 2023 - 14th Int. Conf. and Exh. on 3D Body Scanning and Processing Technologies, Lugano, Switzerland, 17-18 Oct. 2023, #29, https://doi.org/10.15221/23.29.

Title:

Skinning Weights Optimization through Data-Driven Approaches: Determining the Training Dataset

Authors:

Alejandra POLANCO 1, Yoann LAFON 1, Georges BEURIER 1, Junfeng PENG 2, Xuguang WANG 1

1 Univ. Lyon, Univ Gustave Eiffel, Univ Claude Bernard Lyon 1, LBMC UMR_T, Lyon, France;
2 Digital Human Modelling Laboratory, Decathlon Sportslab, Lille, France

Abstract:

Skinning methods that employ skeleton-driven deformation for predicting human skin shapes in different postures are widely utilized in computer graphics applications, but also in ergonomics-related applications for products and clothes development. In these methods, a hierarchical bone structure is used to drive the transformation between postures of the vertices compositing the skin mesh. The vertices are influenced by the underlying bones according to their skinning weights. The skinning weights can be defined from manual painting, purely geometric methods, or solving heat diffusion equations among others. Data-driven approaches use large amounts of data to determine the skinning weighs: 64000 multi-pose and multi-identity body scans were used in [1], 1786 multi-pose meshes from 40 subjects and 3800 multi-identity meshes were used in [2], and 237 meshes from 50 subjects were used in [3]. Even though it’s well-known that numerous training skins are required to produce accurate results, the sensitivity of the results to the amount and type of training data has not been investigated. For this reason, this work studies the influence of the composition of the training dataset (numbers and type of subjects and postures) on the determination of the skinning weights. The quality of the skinning weights was performed by analyzing the predicted skin shapes and 3D scans in the final postures considered as ground truth data. The results of this study can be of interest for the definition of databases used to develop data-driven models to predict shapes in new postures; especially considering the costs associated with data collection campaigns.

Details/PDF/VIDEO:

Abstract: PDF
Presentation: VIDEO
Proceedings: 3DBODY.TECH 2023, 17-18 Oct. 2023, Lugano, Switzerland
Paper id#: 29
DOI: 10.15221/23.29

Copyright notice

© Hometrica Consulting - Dr. Nicola D'Apuzzo, Switzerland, hometrica.ch.
Reproduction of the proceedings or any parts thereof (excluding short quotations for the use in the preparation of reviews and technical and scientific papers) may be made only after obtaining the specific approval of the publisher. The papers appearing in the proceedings reflect the author's opinions. Their inclusion in these publications does not necessary constitute endorsement by the editor or by the publisher. Authors retain all rights to individual papers.

Note: click the + on the top left of the page to open/close the menu.