EmoGen: a methodology for 3D facial expression synthesis and analysis.
generation of arbitrary 3D facial configurations by non-expert users via a genetic algorithm (with a posteriori regularisation)
principled quantitative analysis of generated samples for experimental psychology research
Funding and position: Medical Research Council (MRC), Research Associate
Related research outputs:
Standalone C++ EmoGen tool deployed to the psychologists of Queen Mary University and University College London to enable research.
N. Roubtsova, M. Parsons, N. Binetti, I. Maraschal, E. Viding and D. Cosker, EmoGen: Quantifiable Emotion Generation and Analysis for Experimental Psychology, 2021 ArXiv
Nicola Binetti, Nadejda Roubtsova, Christina Carlisi, Darren Cosker, Essi Viding and Isabelle Mareschal, Genetic algorithms reveal profound individual differences in emotion recognition. Proceedings of the National Academy of Sciences (PNAS) of the United States of America, 119(45), 2022.
Nicola Binetti, Nadejda Roubtsova, Essi Viding, Darren Cosker, Christina Carlisi and Isabelle Mareschal, Individual differences in representation and recognition of facial expressions revealed though a Genetic Algorithm framework. ISRE 2022 (abstract)
Christina Carlisi, Nicola Binetti, Nadejda Roubtsova, Rosie Duffy, Tori Sasaki, Darren Cosker, Isabelle Mareschal and Essi Viding, Understanding the relationship between facial emotion representation and mental health. SAS 2022 (abstract)
Initialisation
Interface
Next generation
Automatic collision correction (regularisation)
Analysis (e.g. genetic algorithm convergence)
December 2016 - January 2019
Scalable Performance-driven Facial Animation
Funding and position: Innovate UK and CAMERA, Research Associate
Related research outputs:
Synthesia , CAMERA (University of Bath) and
Dimension Studio, Innovate UK HARPC: High-Quality Facial Capture Unencumbered by Head Mounted Cameras, industry talk and production test showcase, CVMP 2018.
N.Roubtsova and D.P. Cosker, Scalable Performance-Driven Facial Animation (draft)
Generalisation to non-human actors:
October 2015 - November 2016
Shape and Reflectance Acquisition of Complex Dynamic Scenes
Funding and position: Engineering and Physical Sciences Research Council (EPSRC), Research Fellow
Related research outputs:
N. Roubtsova and J.-Y. Guillemaut, Decoupled Shape and Appearance Acquisition for Photometrically Complex Scenes, short paper,
CVMP 2016 (poster presentation) Best Poster Award
Geometry is estimated using Colour Helmholtz Stereopsis (see below) from just three RGB images per frame and the reflectance model is fitted using six samples per vertex.
Per-pixel chromaticity is calibrated separately.
October 2012 - October 2015
Accurate 3D Reconstruction of Dynamic Scenes with Complex Reflectance Properties
Funding and position: CVSSP, University of Surrey, PhD student
N. Roubtsova, Accurate 3D Reconstruction of Dynamic Scenes with Complex Reflectance Properties, PhD thesis, 2016.
N. Roubtsova and J.-Y. Guillemaut, Bayesian Helmholtz Stereopsis with Integrability Prior, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018.
N. Roubtsova and J.-Y. Guillemaut, Colour Helmholtz Stereopsis for Reconstruction of Dynamic Scenes with Arbitrary Unknown Reflectance, International Journal of Computer Vision, 2017.
N. Roubtsova and J.-Y. Guillemaut, Colour Helmholtz Stereopsis for Reconstruction of Complex Dynamic Scenes, International Conference on 3D Vision, 2014.
N.S. Roubtsova and J.-Y. Guillemaut, A Bayesian Framework for Enhanced Geometric Reconstruction of Complex Objects by Helmholtz Stereopsis, International Conference on Computer Vision Theory and Applications, 2014.
Best Student Paper Award
N. Roubtsova and J.-Y. Guillemaut, Extended Bayesian Helmholtz Stereopsis for Enhanced Geometric Reconstruction of Complex Objects, Springer Lecture Notes.
Computer Vision, Imaging and Computer Graphics - Theory and Applications (VISIGRAPP 2014), Revised Selected Papers, series Communications in Computer and Information Science, 2015.
N. Roubtsova and J.-Y. Guillemaut, Helmholtz Stereopsis Synthetic Dataset , University of Surrey, 2020 (dataset website)
N. Roubtsova and and J.-Y. Guillemaut, Colour Helmholtz Stereopsis Dataset University of Surrey, 2016 (dataset website)
COARSE-TO-FINE BAYESIAN HELMHOLTZ STEREOPSIS WITH INTEGRABILITY PRIOR
Dataset: Doll
Final dense reconstruction and normal map
Dataset: Billiard ball
Final dense reconstruction and normal map
Dataset: Teapot
Final dense reconstruction and normal map
Dataset: Teddy
Final dense reconstruction and normal map
COLOUR HELMHOLTZ STEREOPSIS
Set-up (wavelength multiplexing) and input (3 RGB images)
Reconstruction
Dynamic scenes
Reconstructions (presented untextured with smooth shading) are obtained using the proposed Coarse-to-Fine Bayesian Colour Helmholtz Stereopsis (calibrated)
