Our paper Smoothed Graph Contrastive Learning via Seamless Proximity Integration
with Maysam Behmanesh has been accepted at Learning on Graphs Conference 2024.
I'm currently accepting applications for PhD positions via the ELLIS PhD program portal. Please apply there if you are interested in joining our group.
Souhaib Attaiki a student from our group, just defended his PhD dissertation entitled Robust Deep Learning-based Methods for Non-Rigid Shape Correspondence
. Congratulations to Souhaib!
Our paper GANFusion: Feed-Forward Text-to-3D with Diffusion in GAN Space
with Souhaib Attaiki, Paul Guerrero, Duygu Ceylan and Niloy Mitra has been accepted at WACV 2025.
I'm happy and honored to join Google DeepMind as a Visiting Research Scientist.
Qixing Huang from UT Austin is visiting our group for 3 weeks, as part of his sabbatical.
Our paper DeBaRA: Denoising-Based 3D Room Arrangement Generation
with Léopold Maillard, Nicolas Sereyjol-Garros, and Tom Durand, was accepted at NeurIPS 2024.
Our paper Deformation Recovery: Localized Learning for Detail-Preserving Deformations
with Ramana Sundararaman, Nicolas Donati, Simone Melzi and Etienne Corman was accepted at SIGGRAPH Asia 2024 (Journal Track).
I gave a talk in the NVIDIA Toronto AI Seminar on Convergence of (some) 3D deep learning approaches
.
Gautam Pai, who did a postdoc in our group, just received a position as an Assistant Professor at the University of Twente (UT) - Netherlands. Congratulations to Gautam!
Robin Magnet a student from our group, just defended his PhD dissertation entitled Robust Spectral Methods for Shape Analysis and Deformation Assessment
. Congratulations to Robin!
Our paper To Supervise or Not to Supervise: Understanding and Addressing the Key Challenges of Point Cloud Transfer Learning
with Souhail Hadgi and Lei Li was accepted at ECCV 2024.
Our paper Fine-tuning 3D foundation models for geometric object retrieval
with Jarne Van den Herrewegen, Francis Wyffels and Tom Tourwe was accepted at 3DOR 2024 (Computers & Graphics journal).
Nicolas Donati, who did his PhD in our group was given the runner-up award for the IDIA IP Paris Best PhD. Congratulations to Nicolas!
Our paper on Complex Functional Maps was recognised as a top cited paper in the Computer Graphics Forum journal for the period 2022-2023.
Vincent Mallet, who did a postdoc in our group just received a permanent researcher position at École des Mines de Paris. Congratulations to Vincent!
We have four papers accepted at CVPR 2024:
Back to 3D: Few-Shot 3D Keypoint Detection with Back-Projected 2D Featureswith Thomas Wimmer and Peter Wonka,
Scalable and Simplified Functional Map Learningwith Robin Magnet
Self-Supervised Dual Contouringwith Ramana Sundararaman and Roman Klokov
PoNQ: a Neural QEM-based Mesh Representationwith Nissim Maruani, Pierre Alliez and Mathieu Desbrun
In the new year, we have three new group members: Emery Pierson as a postdoctoral researcher, as well as Diego Gomez (supported by the ERC project VEGA) and Léopold Maillard (as an industrial PhD student in partnership with Dassault Systèmes). In addition, Mazdak Abulnaga (MIT CSAIL and Harvard Medical School) came to visit us for a month as a research collaborator. Welcome to Emery, Diego, Léopold and Maz!
I gave a talk in the GDR RADIA IFM joint meeting, held in Angers. My talk was on Robust General-Purpose Learning on Surfaces and Graphs
.
Two of our papers: RIVQ-VAE: Discrete Rotation-invariant 3D Representation Learning
with Mariem Mezghanni and Malika Boulkenafed; and Unsupervised Representation Learning for Diverse Deformable Shape Collections
with Sara Hahner, Souhaib Attaiki, and Jochen Garcke have been accepted at 3DV 2024.
Our paper Shape Non-rigid Kinematics (SNK): A Zero-Shot Method for Non-Rigid Shape Matching via Unsupervised Functional Map Regularized Reconstruction
with Souhaib Attaiki has been accepted at NeurIPS 2023.
I gave one of the two keynote talks at the Machine Learning for Geometry Workshop, held at the Institut Henri Poincaré in Paris. This event in general had a very exciting set of speakers and talks.
As part of the ELLIS program, I'm currently accepting applications for (fully funded) PhD positions via the ELLIS portal. You can find the information here: https://ellis.eu/news/ellis-phd-program-call-for-applications-2023
Tim Scheller has just joined our group as a starting PhD student, to work on topics related to dynamic 3D shape modelling, reconstruction and generation among other things. Welcome to Tim!
I gave one of three keynotes at Pacific Graphics 2023. My talk was dedicated to giving an overview of functional maps, from their inception to recent breakthroughs.
I'm excited to be giving a keynote at 3DOR 2023 on Efficient, general-purpose feature learning for 3D shape comparison
.
I'm happy to be part of three accepted at ICCV 2023:
SATR: Zero-Shot Semantic Segmentation of 3D Shapeswith Ahmed Abdelreheem, Ivan Skorokhodov, and Peter Wonka (from KAUST);
Spatially and Spectrally Consistent Deep Functional Mapswith Mingze Sun, Shiwei Mao, Puhua Jiang, and Ruqi Huang (from TBSI); and
VoroMesh: Learning Watertight Surface Meshes with Voronoi Diagramswith Nissim Maruani, Roman Klokov, Pierre Alliez, and Mathieu Desbrun
I'm happy and honored to have become a fellow of ELLIS, the European Laboratory for Learning and Intelligent Systems, bringing together top AI researchers in Europe.
Our work on Functional Maps has received the ACM SIGGRAPH 2023 Test-of-Time award.
Etienne Corman, who did his PhD in our group in 2013 - 2016 just won the SMI Young Researcher Award. Congratulations to Etienne!
Our paper ReVISOR: ResUNets with visibility and intensity for structured outlier removal
with Maxime Kirgo, and collaborators from EDF R&D Guillaume Terrasse and Guillaume Thibault was accepted at the ISPRS Journal of Photogrammetry and Remote Sensing. This paper presents a 3D deep learning-based method for structured outlier detection, especially arising from reflections in laser scans of large 3D scenes.
Our paper Assessing Craniofacial Growth and Form Without Landmarks
done in collaboration between our group, including Robin Magnet, and Simone Melzi, as well as a craniofacial surgeon and researcher Roman Khonsari and his group has been accepted in the Journal of Morphology. In this paper we present a novel fully-automatic method for the morphometric analysis of 3D shapes, and use it for detecting and quantifying two craniofacial anomalies: trigonocephaly and metopic ridges, using CT-scans of young children.
I gave an invited talk at the 3D Vision Summer School on Recent learning-based approaches for 3D shape understanding and correspondence
.
Our paper TIDE: Time Derivative Diffusion for Deep Learning on Graphs
with Maximilian Krahn and Maysam Behmanesh has been accepted at ICML 2023. This paper presents an efficient method for learning on graphs that allows long-range communication without oversmoothing, and, at the same time, inherits the full expressive power of local message passing approaches.
Three of our papers have been accepted at CVPR 2023:
Affection: Learning Affective Explanations for Real-World Visual Datawith Panos Achlioptas, Leonidas Guibas, and Sergey Tulyakov,
Generalizable Local Feature Pre-training for Deformable Shape Analysiswith Souhaib Attaiki and Lei Li, and
Understanding and Improving Features Learned in Deep Functional Mapswith Souhaib Attaiki
While not research related, perhaps the most important news of all, our baby daughter Ida was born in February 2023! Her mother and her are both doing well (to the extent possible).
I'm very happy and honored to have just received an ERC Consolidator Grant for my project VEGA: Universal Geometric Transfer Learning. This project will explore ways in which 3D geometric data can be made more amenable to learning, especially in the presence of limited labeled training data. The official announcement is here. As always, I'm looking for motivated PhD and Postdoc candidates!
Two of my outstanding PhD students have successfully defended their PhDs in January 2023:
Robust representations for supervised and unsupervised 3D shape matching, and
Structural and Functional Learning for Industrial Design Automatization,done in partnership with Dassault Systèmes.
Our paper on Functional Maps on Dense Meshes
with Robin Magnet has been accepted for publication at Eurographics 2023. This paper presents a scalable approach for computing correspondences across very dense meshes with theoretical guarantees.
I'm very happy to be part of 3 papers accepted at NeurIPS 2022:
Learning Multi-resolution Functional Maps with Spectral Attention for Robust Shape Matchingwith Lei Li and Nicolas Donati,
Neural Correspondence Prior for Effective Unsupervised Shape Matchingwith Souhaib Attaiki
Reduced Representation of Deformation Fields for Effective Non-rigid Shape Matchingwith Ramana Subramanyam Sundararaman, Riccardo Marin and Emanuele Rodolà
It was a privilege to be a co-organizer of Hi! PARIS Meet Up! on Computer Vision. This ended up being a fantastic event bringing together PhD students and experts from both academia and industry to discuss the new frontiers of Computer Vision research.
Our paper Smooth Non-Rigid Shape Matching via Effective Dirichlet Energy Optimization
with Robin Magnet, Jing Ren and Olga Sorkine-Hornung has won the best paper award at 3DV2022. Congratulations to Robin and Jing!
We have two papers accepted at 3DV 2022: Smooth Non-Rigid Shape Matching via Effective Dirichlet Energy Optimization
(as an oral) with Robin Magnet, Jing Ren and Olga Sorkine-Hornung and SRFeat: Learning Locally Accurate and Globally Consistent Non-Rigid Shape Correspondence
(as a poster) with Lei Li and Souhaib Attaiki. These papers present, respectively, a novel approach for promoting pointwise map smoothness in functional map computations, and a novel learning-based framework that combines the local accuracy of contrastive learning with the global consistency of geometric approaches, for robust non-rigid matching.
Our paper on Implicit field supervision for robust non-rigid shape matching
with Ramana Sundararaman and Gautam Pai has been accepted to ECCV 2022 as an oral. This paper presents a very robust learning-based method for non-rigid shape correspondence, using the neural field shape representation.
Marie-Julie Rakotosaona, who did her PhD in our group, has won the best thesis award from GdR IG-RV, the French association for Computer Graphics, Virtual Reality and Visualization. In addition, she also won the best thesis second prize award from the IDIA department. Congratulations to Marie-Julie!
Our paper on Non-Isometric Shape Matching via Functional Maps on Landmark-Adapted Bases
with Mikhail Panine and Maxime Kirgo has been accepted at Computer Graphics Forum. This paper proposes a method for landmark-preserving shape correspondence with functional maps that aims at as conformal as possible maps (unlike the standard isometry model), without using any descriptors.
Our papers on the Physical Simulation Layer for Accurate 3D Modeling
with Mariem Mezghanni, Théo Bodrito and Malika Boulkenafed; and on Deep Orientation-Aware Functional Maps: Tackling Symmetry Issues in Shape Matching
with Nicolas Donati and Etienne Corman, were accepted at CVPR 2022 as an oral and a poster respectively. The preprints are now on my Publications page.
Our paper PointCleanNet: Learning to Denoise and Remove Outliers from Dense Point Clouds was recognized as a top cited article in the Computer Graphics Forum journal in 2020-2021.
Our paper WSDesc: Weakly Supervised 3D Local Descriptor Learning for Point Cloud Registration
with Lei Li and Hongbo Fu has been accepted in the IEEE Transactions on Visualization and Computer Graphics (TVCG) journal.
Our paper on Spectral Unions of Partial Deformable 3D Shapes
with Luca Moschella, Simone Melzi, Luca Cosmo, Filippo Maggioli, Or Litany, Leonidas Guibas and Emanuele Rodolà has been accepted to Eurographics 2022.
Our paper on DiffusionNet: Discretization Agnostic Learning on Surfaces
with Nicholas Sharp, Souhaib Attaiki and Keenan Crane has been accepted in ACM Trans. on Graph. (TOG). This paper, for which, the code is already available online proposes a new method for learning on surfaces, which is fast, robust and very accurate. Waiting for the TOG acceptance hasn't stopped us from already using it in several other projects!
Our paper on Complex Functional Maps: a Conformal Link Between Tangent Bundles
with Nicolas Donati, Etienne Corman and Simone Melzi has been accepted in the Computer Graphics Forum journal. This work presents a generalization of functional maps that is orientation-aware by using the complex structure of the tangent bundle.
Marie-Julie Rakotosaona, a PhD student from our group, has defended her dissertation on Learning-based representations and methods for 3D shape analysis, manipulation and reconstruction
, and has accepted a research scientist position at Google Research. Congratulations Marie-Julie!
Our DPFM
paper with Souhaib Attaiki and Gautam Pai has won the best paper award at 3DV 2021. Congratulations to Souhaib and Gautam!
Our work entitled DPFM: Deep Partial Functional Maps
with Souhaib Attaiki and Gautam Pai has been accepted at 3DV 2021 as an oral.
I gave a talk on Efficient learning on curved surfaces via diffusion
at the Geometry & Learning from Data workshop organized by BIRS (the Banff center). The video of my talk is now online on the workshop website.
I'm very happy to have two papers accepted at SIGGRAPH Asia 2021: Differentiable Surface Triangulation
with Marie-Julie Rakotosaona, Noam Aigerman, Niloy J. Mitra, and Paul Guerrero, and Intuitive and Efficient Roof Modeling for Reconstruction and Synthesis
with Jing Ren, Biao Zhang, Bojian Wu, Jianqiang Huang, Lubin Fan and Peter Wonka.
Our paper on DWKS : A Local Descriptor of Deformations Between Meshes and Point Clouds
with Robin Magnet has been accepted at ICCV.
Our paper on Spectral Shape Recovery and Analysis Via Data-driven Connections
with Riccardo Marin, Arianna Rampini, Umberto Castellani, Rodolà and Simone Melzi has been accepted to the International Journal of Computer Vision (IJCV).
I gave a talk in the London Geometry and Machine Learning Summer School on Robust learning-based methods for shape correspondence. The talk (among many other amazing talks and materials) is available on the summer school website
Adrien Poulenard, who did a PhD in our group, has won the best thesis award from the IDIA department of IP Paris. He gave a short 10 minute talk on this occasion, to summarize some of this work, which can be found here. Congratulations to Adrien!
Our CVPR paper on Learning Delaunay Surface Elements for Mesh Reconstruction
and the work of Marie-Julie Rakotosaona, a senior PhD student in our group, were featured in the CVPR daily.
Our paper on Discrete Optimization of Shape Matching
with Jing Ren, Simone Melzi and Peter Wonka has been accepted at the Symposium on Geometry Processing (SGP) 2021.
I gave a talk on Robust and Efficient Geometric DL for Non-Rigid Shape Processing, as part of the 3DGV is a virtual seminar series on Geometry Processing and 3D Computer Vision.
I gave a talk in the Sony &inCSL seminar series, on Spectral methods for non-rigid shape comparison
.
I'm very happy to be part of 4 papers accepted at CVPR 2021:
ArtEmis: Affective Language for Visual Artwith Panos Achlioptas, Kilichbek Haydarov, Mohamed Elhoseiny and Leonidas Guibas (oral)
Physically-aware Generative Network for 3D Shape Modelingwith Mariem Mezghanni, Malika Boulkenafed and Andre Lieutier (poster)
Learning Delaunay Surface Elements for Mesh Reconstructionwith Paul Guerrero, Noam Aigerman and Niloy Mitra (oral)
Fast Sinkhorn Filters: Using Matrix Scaling for Non-Rigid Shape Correspondence with Functional Mapswith Gautam Pai, Jing Ren, Peter Wonka and Simone Melzi (poster)
I gave a talk in the TUM AI Lecture Series. The talk was on various ways of modeling and solving matching problems with learning techniques.
Our work on ArtEmis: Affective Language for Art has attracted some media interest, including articles in Forbes Science, New Scientist, HAI, Communications of the ACM, an article in Medium and general public article on the Ecole Polytechnique website, among others.
We are organizing a Tutorial on Inverse Spectral Geometry at Eurographics 2021. This tutorial will overview the recent (very exciting!) developments in recovering geometry from spectral quantities. A preliminary description can be found on the Eurographics website. Please join us if you are interested in how to hear the shape of the drum
.
Our paper Orthogonalized Fourier Polynomials for Signal Approximation and Transfer
with the colleagues from Sapienza University of Rome and ICL has been accepted at Eurographics 2021. The code, prepared by Filippo Maggioli is available here.
Our paper on Geometric analysis of shape variability of lower jaws of prehistoric humans
has been published in the L'Anthropologie. Our first publication in such a venue. It's a step towards making shape comparison more automatic and reproducible in different sciences.
I gave a talk on Efficient Methods for 3D Shape Comparison in the Christmas Colloquium on Computer Vision organised by Skoltech and Samsung. The event was geared towards Russian speakers, but the lectures (including mine) are mostly in English.
Our paper on Wavelet-based Heat Kernel Derivatives: Towards Informative Localized Shape Analysis
has been accepted at Computer Graphics Forum. The code is also available now on this website.
Our paper Instant recovery of shape from spectrum via latent space connections
that was first-authred by Riccardo Marin won the Best Student Paper award at 3DV 2020. Congratulations, Riccardo!
Our paper on Instant recovery of shape from spectrum via latent space connections
has been accepted as an oral to 3DV 2020.
Two of our papers Weakly Supervised Deep Functional Map for Shape Matching
and Correspondence learning via linearly-invariant embedding
have been accepted at NeurIPS 2020.
Our paper MapTree: Recovering Multiple Solutions in the Space of Maps
has been accepted at SIGGRAPH Asia 2020.
Our papers Intrinsic Point Cloud Interpolation via Dual Latent Space Navigation
and PointTriNet: Learned Triangulation of 3D Point Sets
have been accepted at ECCV 2020 as an oral and a poster respectively.
I gave a keynote at SGP 2020 on Efficient methods for 3D shape comparison, matching and interpolation
.
Our paper Deep Geometric Functional Maps: Robust Feature Learning for Shape Correspondence
was one of the 26 papers that received a Best Paper Award Nomination at CVPR 2020 among 5865 submissions to the conference.
Our paper on Consistent ZoomOut: Efficient Spectral Map Synchronization
has been accepted at SGP 2020.
I gave a keynote at SMI 2020 and a course in the summer school Machine Learning for Non-Matrix Data
on 3d Shape Matching and Registration.
Adrien Poulenard, a PhD student from our group, has defended his dissertation on Structures for deep learning and topology optimization of functions on 3D shapes
in the first (and so far only) virtual defense. Congratulations Adrien!
Thibault Lescoat, a PhD student from our group, has defended his dissertation on Geometric operators for 3D modeling using dictionary-based shape representations
. Congratulations Thibault!
Our paper on Geometric Functional Maps: Robust Feature Learning for Shape Correspondence
has been accepted at CVPR 2020.
Our paper on Spectral Mesh Simplification
has been accepted at Eurographics 2020.
My project AIGRETTE: Analyzing Large Scale Geometric Data Collections
has just been selected among the recipients of the Artificial Intelligence Chair in France. I am looking for candidates interested in PhD and Postdoc positions to work on topics related to 3D Machine Learning. Please take a look at the Open Positions page.
Our paper Unsupervised Deep Learning for Structured Shape Matching
was one of the seven papers that received a Best Paper Award Nomination at ICCV 2019 among 4303 submissions to the conference.
Our paper Correspondence-Free Region Localization for Partial Shape Similarity via Hamiltonian Spectrum Alignment
won the best paper award at 3DV 2019.
I gave a talk at the Max Planck Institute for Informatics on recent methods for shape matching with and without learning.
Our paper ZoomOut: Spectral Upsampling for Efficient Shape Correspondence
was accepted at SIGGRAPH Asia 2019.
We have two papers accepted at 3DV 2019: Correspondence-Free Region Localization for Partial Shape Similarity via Hamiltonian Spectrum Alignment
and Effective Rotation-invariant Point CNN with Spherical Harmonics Kernels,
both as orals.
We have two papers accepted at ICCV 2019: OperatorNet: Recovering 3D Shapes From Difference Operators
as a poster and Unsupervised Deep Learning for Structured Shape Matching
as an oral.
Our paper Structured Regularization of Functional Map Computations
has won the Best Paper award honorable mention at SGP 2019.
Two of our papers were accepted at SGP 2019: Structured Regularization of Functional Map Computations
and Limit Shapes – A Tool for Understanding Shape Differences and Variability in 3D Model Collections
. The preprints are now on my Publications page.
Our paper PointCleanNet: Learning to Denoise and Remove Outliers from Dense Point Clouds
has been accepted in the Computer Graphics Forum. The paper, code and data can be found on a dedicated website.
I gave talks on recent methods in non-rigid shape matching at NYU and Stanford.
I gave a talk on non-rigid shape matching at IPAM in UCLA.
Our paper on Spectral Coarsening of Geometric Operators
has been accepted at SIGGRAPH 2019.
I gave a talk on Deep learning-based approaches for 3D shape processing and analysis
at Inria Rhône-Alpes
Our paper on Isospectralization, or how to hear shape, style, and correspondence
has been accepted to CVPR 2019.
I gave a talk at the Point Cloud Analysis seminar in Paris on Deep Learning Approaches for 3D Point Cloud Processing
.
Dorian Nogneng, a PhD student from our group, has defended his doctoral dissertation on Non-rigid correspondences between surfaces embedded in 3D
. Congratulations Dorian!
Our paper on: Functional Characterization of Deformation Fields
has been accepted to the Transactions on Graphics journal and was invited to be presented at SIGGRAPH 2019.
Our paper on: Spectral Measures of Distortion for Change Detection in Dynamic Graphs
has been accepted to the Complex Networks 2018 conference as an Oral paper. The preprint will soon be on my Publications page.
Two of our papers: Multi-directional Geodesic Neural Networks via Equivariant Convolution
and Continuous and Orientation-preserving Correspondences via Functional Maps
have been accepted to SIGGRAPH Asia 2018. Their preprints can now be found on my Publications page.
I co-organized a tutorial on Functional Maps at ECCV 2018 in Munich. Our tutorial covered the basics of the functional maps representation and its applications in Computer Vision. The slides from all the presentations are now available on the dedicated tutorial website
Our paper on Joint Graph Layouts for Visualizing Collections of Segmented Meshes
has been selected as the Spotlight paper for the September 2018 issue of the TVCG journal.
Our paper on Functional Characterization of Deformation Fields
has been accepted with minor revisions to the Transactions on Graphics journal.
I gave a talk at the mini-symposium on Shape processing at the Curves and Surfaces 2018 conference. The abstract for my talk is on the conference website and the slides for my talk are online.
The programs for both the conference and the graduate school have been set for SGP 2018. We have a very exciting set of courses and talks scheduled. You can find all the relevant information on the Conference Website.
Our paper on Topological Function Optimization for Continuous Shape Matching was accepted at SGP 2018. The preprint is now on my Publications page.
We presented the Survey on Data-driven Dictionary-based Methods for 3D Modeling at Eurographics 2018 in Delft. You can find the survey itself along with the bibliography on the dedicated page.
I gave a talk at the Workshop: Imaging and Vision from Theory to Applications in Siegen, Germany. The slides for my talk on Improved Functional Mappings via Product Preservation are now online.
We just released the code for several papers, including
Our survey on Data-driven Dictionary-based Methods for 3D Modeling has been accepted with minor revisions to the State-of-the-Art reports track of Eurographics 2018.
I gave a talk at the Workshop on Flows, mappings and shapes. The slides for my talk on Efficient regularization of functional map computations are now online. You can also find the video recording of the talk on the Newton Institute website.
Our papers on Improved Functional Mappings via Product Preservation, and PCPNET Learning Local Shape Properties from Raw Point Clouds have been accepted to Eurographics 2018. The preprints for both are now on my publications page.
I am looking to hire a postdoctoral researcher and a PhD student to work on a project related to analyzing collections of 3D shapes. Both positions will start in April-September 2018. Please see the following announcement with the description of the positions.
Our paper on Robust Structure-based Shape Correspondence has been conditionally accepted with minor revisions to Computer Graphics Forum. The pre-print is now on arxiv.
I gave a talk at the Workshop on Distance Geometry. The slides for my talk on Recovering the metric from the Laplacian are now online.
Together with Tamy Boubekeur and Pooran Memari, I am co-organizing SGP 2018 that will take place in Paris. The dates have been set and can be found on the conference website that we have created. Hope to see you there!
The pre-print for our ICCV paper on Region-based Correspondence Between 3D Shapes via Spatially Smooth Biclustering together with a code demo and a poster prepared by Matteo Dennito and Simone Melzi are now all linked on my publications page.
I gave a talk at the Geometry Workshop in Obergurgl. The slides for my talk on Some structural properties of functional map computation are now online.
I have just received an ERC Starting Grant for a project on exploring relations in structured data with functional maps. The official announcement is here.
