This website is home to the French research community on shapes analysis, from biomedical imaging to computer assisted design and graphics.
We organize an open seminar in Paris and diffuse announcements related to relevant software tools, conferences and job offers. You may subscribe to the mailing list, feel free to invite newcomers in the field (interns, students, engineers…). You may also consider to subscribe to the calendar feed.
Every three weeks, we organize an in-person seminar in Paris around noon:
Clément Zanolli (Laboratory PACEA, University of Bordeaux, CNRS)
at 12pm
in TBA.
Abstract:
Since the 19th century, the study of human evolution has mostly been based on the study of morphological variation observed in the fossil record. The use of “classic” paleoanthropological approaches using linear measurements and visual descriptions has now moved to the three-dimensional quantification of size and shape thanks to the development of virtual imaging approaches. Since the first use of X-ray microtomography on a hominin tooth in 2004, the field of virtual paleoanthropology has successfully grown and has become state-of-the-art for the study of the external and internal structure of fossil teeth and bones. In parallel, geometric morphometric approaches became a standard method to finely characterize morphological variation and, coupled with statistical methods, improve our understanding of hominin evolution. This presentation will show a few examples of application of 3D imaging techniques that contributed to resolving scientific questions related to hominin diversity and their evolutionary trajectories.
Maxime Taverne (APHP, Necker Enfants Malades, Craniofacial Growth and Form lab)
at 12pm
in TBA.
Abstract:
Maxime will present an overview of his research on the statistics of anatomical shapes, from lizard forearms to human skulls.
Pascal Romon (Université Paris-Est Marne-la-Vallée - LAMA)
at 12pm
in the seminar room of the MAP5 laboratory, 45 rue des Saints-Pères in Paris (7th floor) - detailed path.
Abstract:
The estimation of differential quantities on oriented point cloud is a classical step for many geometry processing tasks in computer graphics and vision. Even if many solutions exist to estimate such quantities, they usually fail at satisfying both a stable estimation with theoretical guarantee, and the efficiency of the associated algorithm. Relying on the notion of corrected curvature measures designed for surfaces, the method introduced in this paper meets both requirements. Given a point of interest and a few nearest neighbours, our method estimates the whole curvature tensor information by generating random triangles within these neighbours and normalising the corrected curvature measures by the corrected area measure. We provide a stability theorem showing that our pointwise curvatures are accurate and convergent, provided the noise in position and normal information has a variance smaller than the radius of neighbourhood. Experiments and comparisons with the state-of-the-art confirm that our approach is more accurate and much faster than alternatives. The method is fully parallelizable, requires only one nearest neighbour request per point of computation, and is trivial to implement.
Siwan Boufadène (Université Gustave Eiffel, LIGM)
at 12pm
in the new Inria Paris centre (48 rue Barrault), in the Leonhard Euler room (A303bis) behind the cafeteria.
Abstract:
In this work, we study the Wasserstein gradient flow of the Riesz energy defined on the space of probability measures. The Riesz kernels define a quadratic functional on the space of measure which is not in general geodesically convex in the Wasserstein geometry, therefore one cannot conclude to global convergence of the Wasserstein gradient flow using standard arguments. Our main result is the exponential convergence of the flow to the minimizer on a closed Riemannian manifold under the condition that the logarithm of the source and target measures are Hölder continuous. To this goal, we first prove that the Polyak-Lojasiewicz inequality is satisfied for sufficiently regular solutions. The key regularity result is the global in-time existence of Hölder solutions if the initial and target data are Hölder continuous, proven either in Euclidean space or on a closed Riemannian manifold. For general measures, we prove using flow interchange techniques that there is no local minima other than the global one for the Coulomb kernel. In fact, we prove that a Lagrangian critical point of the functional for the Coulomb (or Energy distance) kernel is equal to the target everywhere except on singular sets with empty interior. In addition, singular enough measures cannot be critical points. Link to the preprint.
Julien Tierny (CNRS, Sorbonne University, École Polytechnique)
at 12pm
in the new Inria Paris centre (48 rue Barrault), in the Emmy Noether room (A002) on the ground floor behind the auditorium.
Abstract:
Topological Data Analysis (TDA) is a recent area of computer science that focuses on discovering intrinsic structures hidden in data. Based on sound mathematical foundations such as Morse theory and Persistent Homology, TDA enables the robust extraction of the main features of a dataset into stable, concise, and multi-scale descriptors that facilitate data analysis and interpretation. In this talk, I will give an intuitive overview of the main tools used in TDA, and I will illustrate their application in a concrete use-case in computational fluid dynamics. This talk will be illustrated with results produced with the “Topology ToolKit” (TTK), an open-source library (BSD license) that we develop with collaborators to showcase our research. Tutorials for re-producing these experiments are available on the TTK website.
Olivier Couture (CNRS, Sorbonne University)
at 12pm
in the seminar room of the MAP5 laboratory, 45 rue des Saints-Pères in Paris (7th floor) - detailed path.
Abstract:
Ultrasound imaging is a widely used medical technique sensitive to blood flow. However, it is limited by a classic tradeoff between penetration and resolution. In particular, brain imaging and ultrasound angiography are compromised by the skull, which attenuates the acoustic waves, especially those that could provide submillimeter details. Ultrasound localization microscopy (ULM) is a technique that provides very high-resolution depth imaging in tissues. It relies on clinical contrast agents, called microbubbles, which are isolated and localized at the micrometer scale in ultrasound images. Tracking their trajectories can provide blood speeds in the microcirculation. This technique has been demonstrated in the brain (Errico et al. Nature 2015), as well as in the kidney, tumors, peripheral vessels and lymph nodes, in both humans and animals (see the review in Couture et al. IEEE UFFC 2018, Christensen-Jeffries et al. UMB 2020 or Song et al. “Super-resolution ultrasound microvascular imaging: Is it ready for clinical use?” 2023). During this presentation, we will describe recent advances in ULM. In particular, we will show the three-dimensional reconstruction of the rat brain hemodynamics at the micrometer scale. A full-brain angiography at this scale will also be presented. We will discuss its application for the differentiation of strokes and its transfer to the clinical domain through intermediate experiments on the sheep brain and two-dimensional ULM with clinical ultrasound devices. All these developments aim to create a new imaging tool for stroke patients in intensive care, potentially reducing treatment time for ischemic patients. We will also describe recent developments in kidney imaging, where ULM offers new perspectives on the functional units of the organ, namely the nephron and its glomeruli, in both humans and animals. In the future, ULM could enable depth angiography at a scale currently only achievable through invasive modalities.
Jean Feydy (Inria Paris, HeKA team)
at 12pm
in the seminar room of the MAP5 laboratory, 45 rue des Saints-Pères in Paris (7th floor) - detailed path.
Links:
video
slides
Abstract:
This will be a tutorial on the rendering of 3D data for scientific publications. We will start from simple practical questions: how to load a 3D mesh, a vector field, a volumetric image. Then, I will explain how to go from “plain” to “gorgeous” graphics with shadows, textures, animations or physical material models. Finally, I will discuss some advanced tricks to improve the legibility of complex 3D shapes such as brain vessel trees or volumetric CT scans. I will rely on the Visualization ToolKit, a stable and versatile engine that can be used interactively with Paraview and as a substitute for Matplotlib with PyVista and Vedo.
Thérèse E. Malliavin (CNRS, Laboratoire de Physique et Chimie Théoriques at the University of Lorraine)
at 12pm
in the Turing room of the MAP5 laboratory, 45 rue des Saints-Pères in Paris (7th floor) - detailed path.
Links:
video
Abstract:
Structural bioinformatics plays an increasingly important role in the prediction and calculation of biopolymer structures, leading to a better definition of their function. In addition, intrinsically disordered proteins (IDP) and intrinsically disordered regions (IDR) are at the center of numerous regulation and control pathways in the cell, and attract consequently extreme interest nowadays in structural biology. The optimization problem that arises for protein structure determination is more complex for such objects as the convergence criterion is more difficult to set up and the size of the conformational space is a obstacle to exhaustive exploration. The threading-augmented interval Branch-and-Prune (TAiBP), based on a reformulation of the Distance Geometry Problem (DGP), provides a theoretical frame for the fast generation of protein conformations, avoiding the combinatorial explosion of the Branch-and-Prune approach due to exponential complexity. (1, 2) The TAiBP pipeline was applied to a several disordered proteins (3, 4, 5) and a Gaussian mixture approach RamaMix has been developed for the quantification of IDP conformations from the global distributions of backbone torsion angles. The presentation will be oriented to propose possible connections of the structural bioinformatics to the use of methods coming from optimal transport approaches.
Thomas Pierron (ENS Paris-Saclay, Centre Borelli)
at 12pm
in Jussieu - room 15-16.309 (between towers 15 and 16, on the 3rd floor).
Abstract:
This talk will be divided in two parts. First we will study the extension of LDDMM framework to continuous actions of right half-Lie groups equipped with a right invariant sub-Riemannian metric. This generalizes work from Arguillere on the sub-Riemannian geometry of Sobolev diffeomorphisms and their applications to shape analysis. In particular, it allows to introduce news co-adjoint variables and study new dynamics. We will discuss on some optimization problems that this framework raises with some examples of half-Lie group actions
Gautam Pai (Eindhoven University of Technology)
at 12pm
in Jussieu - room 15-16.309 (between towers 15 and 16, on the 3rd floor).
Links:
video
Abstract:
The roto-translation group SE(2) has been of active interest in image analysis due to methods that lift the image data to multi-orientation representations defined in this Lie group. This has led to impactful applications of crossing-preserving flows for image de-noising, geodesic tracking, and roto-translation equivariant deep learning. In this talk, I will enumerate a computational framework for optimal transportation over Lie groups, with a special focus on SE(2). I will describe several theoretical aspects such as the non-optimality of group actions as transport maps, invariance and equivariance of optimal transport, and the quality of the entropic-regularized optimal transport plan using geodesic distance approximations. Finally, I will illustrate a Sinkhorn-like algorithm that can be efficiently implemented using fast and accurate distance approximations of the Lie group and GPU-friendly group convolutions. We report advancements with the experiments on 1) 2D shape/image barycenters, 2) interpolation of planar orientation fields, and 3) Wasserstein gradient flows on SE(2). We observe that our framework of lifting images to SE(2) and optimal transport with left-invariant anisotropic metrics leads to equivariant transport along dominant contours and salient line structures in the image and leads to meaningful interpolations compared to their counterparts on R^2. Link to the preprint.
Open discussion (collective)
at 12pm
in Jussieu - room 15-16.309.
Abstract:
For this end-of-year seminar, we will have an open discussion around optimization methods for shape analysis, built around small presentations.
Philippe Moireau (Inria Saclay and École Polytechnique, MΞDISIM team)
at 12pm
in Jussieu - room 15-16.101.
Links:
video
Abstract:
Data assimilation aims to couple dynamic models of physical systems with available measurements in order to propose a case-specific, true digital twin of the observed physics. This field, originally developed in the 1970s for accurate predictions in environmental sciences, is now being extended to various fields, including engineering and life sciences. In this talk, we propose to adapt classical data assimilation formulations to deal with shape data defined in adapted space manifolds. We will present configurations where the state space in which the dynamics lives is a space manifold, or configurations where the partial observations are defined in a manifold, so that we can adapt our strategy to all cases of shape data assimilation. Finally, our approach will be illustrated by the assimilation of models taken from environmental sciences - in particular wildfire models - or from life sciences - in particular cardiovascular models.
Bruno Lévy (Inria Saclay, ParMA team)
at 12pm
in Jussieu - room 15-25.328: between towers 15 and 25, on the 3rd floor.
Links:
video
Abstract:
In this presentation, I’ll describe a numerical simulation method for free-surface fluids. I will start by giving an intuitive understanding of the physical phenomena involved in fluid dynamics, pressure, viscosity and surface tension. Then I will detail the numerical simulation method, based on the Gallouet-Mérigot numerical scheme, that describes the fluid as a set of cells, that can deform, but that keep a constant volume, and that follow the motion of the fluid (Lagrangian method). The constant volume constraint takes the form of a partial semi-discrete optimal transport. I will present the geometric and numerical aspects of this optimal transport problem.
Josua Sassen (École Normale Supérieure Paris-Saclay)
at 12pm
in Jussieu - room 15-16.309: between towers 15 and 16, on the 3rd floor.
Links:
video
Abstract:
Josua will present his work on surface segmentation and design with applications to architecture, mechanical engineering and graphics.
Robin Magnet (GeoVic team, LIX, Ecole Polytechnique)
at 12pm
in Jussieu - room 15-16.105 for lunch at noon, 15-16.413 for the talk at 1pm.
Links:
video
Abstract:
Robin will present his work on functional maps, including his pyFM library.
Louis Pujol (Université Paris-Cité, HeKA team)
at 12pm
in the seminar room of the MAP5 laboratory, 45 rue des Saints-Pères in Paris (7th floor) - detailed path.
Links:
video
slides
Abstract:
Louis will present the first (beta) release of the scikit-shapes library. The talk will be followed by an informal lab session - feel free to bring your laptop! Links: tutorial, GitHub, documentation, one-page gallery.
Slavica Jonic (IMPMC – UMR 7590 CNRS, Sorbonne University)
at 12pm
in Jussieu - room 15-16.309: between towers 15 and 16, on the 3rd floor.
Links:
video
Abstract:
The elucidation of different conformations of biomolecular complexes is the key for understanding the molecular mechanisms behind the biological functions of the complexes and the key to novel drug discovery. Single particle analysis (SPA) is a data collection and analysis technique of cryo electron microscopy (cryo-EM) that allows 3D reconstruction of multiple conformations of purified biomolecular complexes from their 2D images. Another cryo-EM technique is cryo electron tomography (cryo-ET), which allows obtaining multiple conformations of the complexes in their cellular environment. Currently, one of the most important and challenging open problems in the cryo-EM field is how to obtain the entire, continuous conformational landscapes of the biomolecules from the data. Consequently, since a few years, methods development to solve this problem has been a very active research area. My group is developing such methods, by combining image processing, molecular dynamics simulations, and deep learning approaches. These methods are made available publicly via our open-source, ContinuousFlex software package, which can be run as a plugin of the broadly used Scipion software package. In this talk, I will present some of our most important achievements regarding this work.
Fleur Gaudfernau (Université de Paris, HeKA team)
at 12pm
in Jussieu - room 15-25.328: between towers 15 and 25, on the 3rd floor.
Links:
video
Abstract:
Fleur will present her PhD thesis work, done under the supervision of Stéphanie Allassonnière and Erwan Le Pennec.
Elodie Maignant (Inria Sophia and École Normale Supérieure Paris-Saclay)
at 12pm
in Jussieu - room 15-16.309: between towers 15 and 16, on the 3rd floor.
Links:
video
Abstract:
Elodie will present her PhD thesis work, done under the supervision of Xavier Pennec and Alain Trouvé.
Barbara Gris (CNRS, Laboratoire Jacques-Louis Lions)
at 10am
in Jussieu - room 15-16.309: between towers 15 and 16, on the 3rd floor.
Links:
video
Abstract:
Barbara will present an optimal control framework to perform statistical shape analysis with plausible biological priors.
Jean Feydy (Inria Paris, HeKA team)
at 9am
in Jussieu - room 15-16.309: between towers 15 and 16, on the 3rd floor.
Links:
video
slides
Abstract:
Jean will present modern optimal transport algorithms, articulating the difference between simple, solved problems in dimension 2 or 3 and NP-hard problems in high dimension.
Josua Sassen (École Normale Supérieure Paris-Saclay)
at 12:30pm
in Jussieu - room 15-25.328: between towers 15 and 25, on the 3rd floor.
Links:
video
Abstract:
Josua is starting a PostDoc in Paris and will introduce himself with a collection of results from his work in Bonn.
Théo Dumont (Université Gustave Eiffel)
at 12pm
in Seminar room of the MAP5 laboratory, 45 rue des Saints-Pères in Paris (7th floor) - detailed path.
Links:
video
Abstract:
Théo will present new results on the structure of the Wasserstein and Gromov-Wasserstein metrics.
Louis Pujol and others (Université Paris-Cité)
at 12pm
in PariSanté Campus, 10 Rue d’Oradour-sur-Glane. Métro Balard or Porte de Versailles. Seminar room of the HeKA team (3rd floor, South corridor).
Abstract:
We will have an open chat on multiscale implementations of shape analysis methods, from both practical and theoretical perspectives. You may prepare up to 5 slides to present a method or implementation that you would like to discuss.
Anna Song (Imperial College London and The Francis Crick Institute)
at 12pm
in Jussieu, room 15-16.309 (i.e. in the corridor that links column 15 to column 16, on the third floor). This is the seminar room of the Laboratoire Jacques-Louis Lions.
Links:
video
slides
Abstract:
Anna will present the results of 4 years of work on 3D shape textures. This includes a new generative model based on curvature minimization, contributions to topological data analysis in 3D and a study of the microscopic structure of bone marrow vessels.
François-Xavier Vialard (Université Gustave Eiffel)
at 12:30pm
in Inria Paris centre, 2 rue Simone Iff, A building, ground floor, Emmy Noether room (Startup Studio entrance).
Links:
slides
Abstract:
François-Xavier will present a collection of open mathematical questions related to diffeomorphic registration and residual networks.
Vincent Mallet (École Polytechnique, CNRS)
at 13pm
Links:
video
slides
Abstract:
Vincent will present his work on the representation of RNA strands as 2.5D graphs, proteins as surfaces and antibody detection on Cryo-EM images.
Guillaume Sérieys (Université Paris-Cité, MAP5)
at 13:30pm
Links:
slides
Abstract:
Guillaume will present his work on the generalization of continuous metamorphoses to images with values in spaces of probability distributions or diffusion tensors.
Thomas Pierron (ENS, Centre Borelli)
at 14pm
Links:
video
slides
Abstract:
Thomas will present his work on a hierarchical, multi-scale model for diffeomorphic shape registration.
Louis Pujol (Université Paris-Cité)
at 12:30pm
in the seminar room of the MAP5 laboratory, 45 rue des Saints-Pères in Paris (7th floor) - detailed path.
Links:
video
code
To register to shapes@inria.fr, please:
A calendar feed of the seminar is available for subscription. Simply add this ical link to your calendar app.