Shanghuan Liu, Shaoyan Gai, Feipeng Da, Fazal Waris
{"title":"Geometry-aware 3D pose transfer using transformer autoencoder","authors":"Shanghuan Liu, Shaoyan Gai, Feipeng Da, Fazal Waris","doi":"10.1007/s41095-023-0379-8","DOIUrl":null,"url":null,"abstract":"<p>3D pose transfer over unorganized point clouds is a challenging generation task, which transfers a source’s pose to a target shape and keeps the target’s identity. Recent deep models have learned deformations and used the target’s identity as a style to modulate the combined features of two shapes or the aligned vertices of the source shape. However, all operations in these models are point-wise and independent and ignore the geometric information on the surface and structure of the input shapes. This disadvantage severely limits the generation and generalization capabilities. In this study, we propose a geometry-aware method based on a novel transformer autoencoder to solve this problem. An efficient self-attention mechanism, that is, cross-covariance attention, was utilized across our framework to perceive the correlations between points at different distances. Specifically, the transformer encoder extracts the target shape’s local geometry details for identity attributes and the source shape’s global geometry structure for pose information. Our transformer decoder efficiently learns deformations and recovers identity properties by fusing and decoding the extracted features in a geometry attentional manner, which does not require corresponding information or modulation steps. The experiments demonstrated that the geometry-aware method achieved state-of-the-art performance in a 3D pose transfer task. The implementation code and data are available at https://github.com/SEULSH/Geometry-Aware-3D-Pose-Transfer-Using-Transformer-Autoencoder.\n</p>","PeriodicalId":37301,"journal":{"name":"Computational Visual Media","volume":null,"pages":null},"PeriodicalIF":17.3000,"publicationDate":"2024-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computational Visual Media","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1007/s41095-023-0379-8","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, SOFTWARE ENGINEERING","Score":null,"Total":0}
引用次数: 0
Abstract
3D pose transfer over unorganized point clouds is a challenging generation task, which transfers a source’s pose to a target shape and keeps the target’s identity. Recent deep models have learned deformations and used the target’s identity as a style to modulate the combined features of two shapes or the aligned vertices of the source shape. However, all operations in these models are point-wise and independent and ignore the geometric information on the surface and structure of the input shapes. This disadvantage severely limits the generation and generalization capabilities. In this study, we propose a geometry-aware method based on a novel transformer autoencoder to solve this problem. An efficient self-attention mechanism, that is, cross-covariance attention, was utilized across our framework to perceive the correlations between points at different distances. Specifically, the transformer encoder extracts the target shape’s local geometry details for identity attributes and the source shape’s global geometry structure for pose information. Our transformer decoder efficiently learns deformations and recovers identity properties by fusing and decoding the extracted features in a geometry attentional manner, which does not require corresponding information or modulation steps. The experiments demonstrated that the geometry-aware method achieved state-of-the-art performance in a 3D pose transfer task. The implementation code and data are available at https://github.com/SEULSH/Geometry-Aware-3D-Pose-Transfer-Using-Transformer-Autoencoder.
期刊介绍:
Computational Visual Media is a peer-reviewed open access journal. It publishes original high-quality research papers and significant review articles on novel ideas, methods, and systems relevant to visual media.
Computational Visual Media publishes articles that focus on, but are not limited to, the following areas:
• Editing and composition of visual media
• Geometric computing for images and video
• Geometry modeling and processing
• Machine learning for visual media
• Physically based animation
• Realistic rendering
• Recognition and understanding of visual media
• Visual computing for robotics
• Visualization and visual analytics
Other interdisciplinary research into visual media that combines aspects of computer graphics, computer vision, image and video processing, geometric computing, and machine learning is also within the journal''s scope.
This is an open access journal, published quarterly by Tsinghua University Press and Springer. The open access fees (article-processing charges) are fully sponsored by Tsinghua University, China. Authors can publish in the journal without any additional charges.