Imperial College London
Alternate

ProfessorAbhijeetGhosh

Faculty of Engineering, Department of Computing

Professor of Graphics and Imaging
 
 
 
//

Contact

 

+44 (0)20 7594 8351abhijeet.ghosh Website

 
 
//

Location

 

376Huxley BuildingSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@unpublished{Lattas:2020,
author = {Lattas, A and Moschoglou, S and Gecer, B and Ploumpis, S and Triantafyllou, V and Ghosh, A and Zafeiriou, S},
publisher = {arXiv},
title = {AvatarMe: realistically renderable 3D facial reconstruction "in-the-wild"},
url = {http://arxiv.org/abs/2003.13845v1},
year = {2020}
}
Download

RIS format (EndNote, RefMan)

TY  - UNPB
AB  - Over the last years, with the advent of Generative Adversarial Networks(GANs), many face analysis tasks have accomplished astounding performance, withapplications including, but not limited to, face generation and 3D facereconstruction from a single "in-the-wild" image. Nevertheless, to the best ofour knowledge, there is no method which can produce high-resolutionphotorealistic 3D faces from "in-the-wild" images and this can be attributed tothe: (a) scarcity of available data for training, and (b) lack of robustmethodologies that can successfully be applied on very high-resolution data. Inthis paper, we introduce AvatarMe, the first method that is able to reconstructphotorealistic 3D faces from a single "in-the-wild" image with an increasinglevel of detail. To achieve this, we capture a large dataset of facial shapeand reflectance and build on a state-of-the-art 3D texture and shapereconstruction method and successively refine its results, while generating theper-pixel diffuse and specular components that are required for realisticrendering. As we demonstrate in a series of qualitative and quantitativeexperiments, AvatarMe outperforms the existing arts by a significant margin andreconstructs authentic, 4K by 6K-resolution 3D faces from a singlelow-resolution image that, for the first time, bridges the uncanny valley.
AU  - Lattas,A
AU  - Moschoglou,S
AU  - Gecer,B
AU  - Ploumpis,S
AU  - Triantafyllou,V
AU  - Ghosh,A
AU  - Zafeiriou,S
PB  - arXiv
PY  - 2020///
TI  - AvatarMe: realistically renderable 3D facial reconstruction "in-the-wild"
UR  - http://arxiv.org/abs/2003.13845v1
UR  - http://hdl.handle.net/10044/1/77830
ER  -
Download