InstantGeoAvatar: Effective Geometry and Appearance Modeling of Animatable Avatars from Monocular Video

Bibliographic Details
Title: InstantGeoAvatar: Effective Geometry and Appearance Modeling of Animatable Avatars from Monocular Video
Authors: Alvaro Budria, Adrian Lopez-Rodriguez, Òscar Lorente, Francesc Moreno-Noguer
Contributors: Agencia Estatal de Investigación (España), Ministerio de Ciencia, Innovación y Universidades (España), Ministerio de Ciencia e Innovación (España), European Commission, Moreno-Noguer, Francesc [0000-0002-8640-684X], Consejo Superior de Investigaciones Científicas [https://ror.org/02gfc7t72]
Source: Lecture Notes in Computer Science ISBN: 9789819609598
UPCommons. Portal del coneixement obert de la UPC
Universitat Politècnica de Catalunya (UPC)
Digital.CSIC. Repositorio Institucional del CSIC
Consejo Superior de Investigaciones Científicas (CSIC)
Publication Status: Preprint
Publisher Information: Springer Nature Singapore, 2024.
Publication Year: 2024
Subject Terms: FOS: Computer and information sciences, Àrees temàtiques de la UPC::Informàtica::Sistemes d'informació::Interacció home-màquina, Computer Vision and Pattern Recognition (cs.CV), Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Aprenentatge automàtic, Computer Science - Computer Vision and Pattern Recognition, Graphics (cs.GR), 3D Computer Vision, Clothed Human Modeling, Human avatars, Computer Science - Graphics, Neural radiance fields, Human Avatars, Clothed human modeling, 3D computer vision, Neural Radiance Fields
Description: We present InstantGeoAvatar, a method for efficient and effective learning from monocular video of detailed 3D geometry and appearance of animatable implicit human avatars. Our key observation is that the optimization of a hash grid encoding to represent a signed distance function (SDF) of the human subject is fraught with instabilities and bad local minima. We thus propose a principled geometry-aware SDF regularization scheme that seamlessly fits into the volume rendering pipeline and adds negligible computational overhead. Our regularization scheme significantly outperforms previous approaches for training SDFs on hash grids. We obtain competitive results in geometry reconstruction and novel view synthesis in as little as five minutes of training time, a significant reduction from the several hours required by previous work. InstantGeoAvatar represents a significant leap forward towards achieving interactive reconstruction of virtual avatars.
Accepted as poster to Asian Conference on Computer Vison (ACCV 2024)
Document Type: Part of book or chapter of book
Article
Conference object
File Description: application/pdf
Language: English
DOI: 10.1007/978-981-96-0960-4_16
DOI: 10.48550/arxiv.2411.01512
DOI: 10.13039/501100000780
DOI: 10.13039/501100004837
DOI: 10.13039/501100011033
Access URL: http://arxiv.org/abs/2411.01512
http://hdl.handle.net/10261/387911
https://api.elsevier.com/content/abstract/scopus_id/85212964933
https://hdl.handle.net/2117/433373
https://doi.org/10.1007/978-981-96-0960-4_16
Rights: Springer Nature TDM
arXiv Non-Exclusive Distribution
Accession Number: edsair.doi.dedup.....be56df04015edfefbaa67ecd8b3d94fb
Database: OpenAIRE
Description
DOI:10.1007/978-981-96-0960-4_16