| Title: |
Generative Articulated Neural Fields to multiple identities and poses |
| Authors: |
Loranchet, Guillaume; Hellier, Pierre; Schnitzler, Francois; Boukhayma, Adnane; Regateiro, Joao; Multon, Franck |
| Contributors: |
Biomécanique et Méthodes Numériques pour l’analyse de l’interaction humain-systèmes hors du laboratoire (ComBO); Université de Rennes 2 (UR2)-Centre Inria de l'Université de Rennes; Institut National de Recherche en Informatique et en Automatique (Inria)-Institut National de Recherche en Informatique et en Automatique (Inria)-RÉALITÉ VIRTUELLE, HUMAINS VIRTUELS, INTERACTIONS ET ROBOTIQUE (IRISA-D5); Institut de Recherche en Informatique et Systèmes Aléatoires (IRISA); Université de Rennes (UR)-Institut National des Sciences Appliquées - Rennes (INSA Rennes); Institut National des Sciences Appliquées (INSA)-Institut National des Sciences Appliquées (INSA)-Université de Bretagne Sud (UBS)-École normale supérieure - Rennes (ENS Rennes)-Institut National de Recherche en Informatique et en Automatique (Inria)-CentraleSupélec-Centre National de la Recherche Scientifique (CNRS)-IMT Atlantique (IMT Atlantique); Institut Mines-Télécom Paris (IMT)-Institut Mines-Télécom Paris (IMT)-Université de Rennes (UR)-Institut National des Sciences Appliquées - Rennes (INSA Rennes); Institut Mines-Télécom Paris (IMT)-Institut Mines-Télécom Paris (IMT)-Institut de Recherche en Informatique et Systèmes Aléatoires (IRISA); Institut National des Sciences Appliquées (INSA)-Institut National des Sciences Appliquées (INSA)-Université de Bretagne Sud (UBS)-École normale supérieure - Rennes (ENS Rennes)-CentraleSupélec-Centre National de la Recherche Scientifique (CNRS)-IMT Atlantique (IMT Atlantique); Institut Mines-Télécom Paris (IMT)-Institut Mines-Télécom Paris (IMT); InterDigital R&D France Cesson-Sévigné; Defi Inria Ys.AI; Quentin Avril; Ludovic Hoyet; Franck Multon; Victoria Abrevaya |
| Source: |
NGA 2025 - Next Generation Avatar Workshop ; https://hal.science/hal-04978097 ; NGA 2025 - Next Generation Avatar Workshop, Quentin Avril; Ludovic Hoyet; Franck Multon; Victoria Abrevaya, Mar 2025, Saint-Malo, France. pp.1-5 |
| Publisher Information: |
CCSD |
| Publication Year: |
2025 |
| Collection: |
Archive Ouverte de l'Université Rennes (HAL) |
| Subject Terms: |
Implicit representation; avatar simulation; multiple identities; fine tuning; AI; character animation; [INFO.INFO-AI]Computer Science [cs]/Artificial Intelligence [cs.AI]; [INFO.INFO-GR]Computer Science [cs]/Graphics [cs.GR] |
| Subject Geographic: |
Saint-Malo; France |
| Description: |
Workshop hosted by IEEE VR2025 conference ; International audience ; The creation of realistic avatars in motion has been the hot-topic topic in academia and creative industry. Recent advances in deep learning and implicit representations have opened new avenues of research, especially to enhance the details of the avatars. However, these approaches generally lead to design large deep learning architectures dedicated to a specific identity shape and pose. Some previous works proposed to adapt the method to handle various poses for a given identity, but they are trained only on one identity. In VR applications, training this type of network for a new user, with a new shape, would require collecting many 3D geometric examples of this user in various poses. Moreover, for distant and massively populated virtual environments, it would also require to transfer one network for each user, which would require huge memory and network usage. This paper proposes an improvement over the state-of-the-art implicit SNARF and Fast-SNARF methods to permit generalization to novel motions and shape identities. Our result show that conditioning a light version of Fast-SNARF with a compact identity latent code enables to handle multiple shape identities while preserving the accuracy of the geometric reconstruction. Extrapolation to identity shapes that have never been seen during training is also possible, but with a little decrease of the reconstruction quality. Hence, to use his/her avatar in VR, a user simply has to provide to the application a compact identity latent code (obtained on a unique 3D example). |
| Document Type: |
conference object |
| Language: |
English |
| Availability: |
https://hal.science/hal-04978097; https://hal.science/hal-04978097v1/document; https://hal.science/hal-04978097v1/file/Generative%20Articulated%20Neural%20Fields%20to%20multiple%20identities%20and%20poses_NGA_2025.pdf |
| Rights: |
https://creativecommons.org/licenses/by-nd/4.0/ ; info:eu-repo/semantics/OpenAccess |
| Accession Number: |
edsbas.E72C351B |
| Database: |
BASE |