Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : ICube, Université de Strasbourg, CNRS
Durée : 3 years
Contact : remi.allegre@unistra.fr
Date limite de publication : 2023-07-30

Contexte :
The aim of this Doctoral thesis is to develop an approach to reconstruct 3D+t (i.e. temporally-consistent) mesh models of growing plants suitable for accurate measurements at fine scales.

Deadline for application: May 23, 2023

Sujet :
Host team: IGG (Computer Graphics and Geometry Group), ICube laboratory

Advisor: Franck Hétroy-Wheeler, Professor in Computer Science (hetroywheeler AT unistra.fr)

Co-advisor: Rémi Allègre, Associate Professor in Computer Science (remi.allegre AT unistra.fr)

Starting date: October 2023

Keywords: Computer Vision, Computer Graphics, Image Processing, Data Science

Description: This doctoral thesis position is proposed in the context of a research project with biophysicists from the University Paris Diderot and ENS Lyon. This project aims at modeling plant growth movements during leaf development and understanding the underlying physical and biological mechanisms at play. In this context, measurements of both plant movements and magnitude of local growth are required. This is currently achieved with the help of photogrammetry only at a coarse scale, considering small sets of markers painted on the leaves. A key challenge of this project is to develop an approach to reconstruct 3D+t (i.e. temporally-consistent) mesh models of growing plants suitable for accurate measurements at fine scales, which involves both high-resolution reconstruction and point-to-point correspondences issues. The goal of this thesis is to address this challenge following a three-part approach: 1) the estimation of optical and scene flows from photographs for fine-scale correspondences between time steps, 2) the combination of different acquisition modalities (photogrammetry, laser scanning and structured light scanning) for high-resolution 3D reconstruction, and 3) the definition of either fine-scale statistical geometric templates for leaves or a neural network architecture for shape interpolation. The developed models and methods will rely on recent machine learning techniques. Several datasets of photographs and 3D reconstructions of growing plants will be provided.

A detailed version of the proposal including bibliography is available at the following address:
https://seafile.unistra.fr/f/93cc5483d1514e3a9b0c/

Profil du candidat :
Desired skills:
– Computer Vision, and/or Computer Graphics or Image Processing, or Data Science
– Basic skills in machine and deep learning

Formation et compétences requises :
M2 or Engineering School degree in Computer Science

Adresse d’emploi :
Illkirch (Strasbourg area)