| My research topics focus on the geometric modeling and analysis of shapes, objects and scenes that exhibit structural regularities, typically man-made objects and urban scenes. |
|
City modeling |
 |
City modeling is a very challenging problem with geometrically different objects, acquisition constraints and scalability issues. City modeling requires analyzing not only the geometry, but also the semantics and structure of objects.
Urban objects of interest include building, vegetation and road mainly. I investigate the design of scalable methods from airborne and satellite data, with a special focus on reconstructing buildings.
Main contributions: Wireframe regularization and extrusion from Lidar [CVPRW24] Polygonal partition extrusion for LOD1 reconstruction [3DGEO19] Learning approach to evaluate city models [PERS19] LOD generation by 3D discrete arrangement [TOG16] Voronoi cell fusion from satellite stereo images [ECCV16] Planimetric arrangement from Lidar [ICCV11, IJCV12] 3D-block assembling from DSM [CVPR08, PAMI10] Roof skeletization [JPRS08] |
|
Data structures and algorithms for piecewise planar geometry |
 |
Space partitioning data structures play a central role in geometric modeling: they are the interface between unorganized data measurements, e.g., 3D point clouds, and standardized algorithms. While freeform geometry benefits from efficient space partitioning data structures, e.g., Delaunay triangulations, piecewise-planar geometry can only rely upon plane arrangements.
I am interested in developping mechanisms to build plane arrangements in a more concise and efficiently way than the standard techniques based on exhaustive plane slicing operations. I am also interested in exploiting these data structures in application scenarios such as low-poly mesh reconstruction.
Main contributions: Adaptive plane arrangement [ECCV24] Mesh repairing with 3D kinetic data structures [JPRS22] Simplification of polygonal partitions[CGF22] 3D Kinetic data structure [TOG20] Reconstruction by a connect-and-slice approach [CVPR20] |
|
Geometric analysis of images |
 |
Extracting objects and structures contained in images by vector representations, e.g., the silhouette of solid objects by polygons or line structures by planar graphs, has been a long-standing problem in Computer Vision.
Such vector representations are compact, editable and can basically approximate well any chain of pixels. I proposed a series of works for producing such representations automatically, mostly inspired by geometry processing concepts.
Main contributions: Generic line-segment detector [ECCV24] Approximating shapes in images with polygons [CVPR20] Polygonal partitioning by kinetic framework [CVPR18] Image partitioning by Voronoi diagram [CVPR15] |
|
Geometric shape detection |
 |
The goal of shape detection consists in turning a large amount of sampling data into a higher level representation based on simple geometric shapes. Shape detection is typically used as a prior step in a large variety of tasks ranging
from surface reconstruction to data registration. I am particularly interested in geometric regularities and structural scale-spaces that govern man-made objects.
Main contributions: Approximation of freeform geometry with Planar shapes [CVPR22] Planar shape detection at structural scales [CVPR18] Planar shape detection and regularization in tandem [CGF15] |
|
Spatial point process |
 |
Spatial point processes describe random configurations of points in a continuous bounded domain. By assigning a parametric object to each point, these stochastic geometry models
become a powerful tool to extract objects with complex spatial interactions from images and 3D data. I am particularly interested in extending point processes to sample more complex structures as triangulations and planar graphs. I am also interested by the design of efficient and scalable samplers for large scale scenes.
Main contributions: Delaunay point processes [PAMI20] Model to extract linear tree structures in remote sensing images[JPRS17] Point process to sample planar graphs [CVPR13] Monte Carlo sampler operating in parallel on large scenes [ECCV12, IJCV14] Jump-Diffusion sampler for multiple types of parametric objects [BMVC08, PAMI10] Model to reconstruct Lidar waveforms [IP10] |
|
Low-power computer vision |
 |
Low power computer vision has emerged in the late 2010’s as a promising research field, in rupture with popular deep learning models based on powerful hardware resources. In this field, accuracy is not the only quest: power/energy, memory footprint and model size are also important objectives. I investigate the design of frugal deep learning architectures and compact object representations for onboard processing of satellite images
Main contributions: Single-Shot End-to-end Road Graph Extraction [CVPRW22] Scanner Neural Network [IGARSS22] Low-power neural networks for onboard satellite image classification [ICCVW19] |
|
Surface reconstruction |
 |
Surface reconstruction is a classical problem in computer vision and
computational geometry. The goal is to interpolate input points by a surface mesh. Input points mainly come from Multi-View Stereo
imagery and Laser scans. My research directions mainly focus on the formulation of new surface quality measures combining different criteria as geometric accuracy, representation compexity,
and structural constraints. I am also interested in piecewise-planar and hybrid representations.
Main contributions: Reconstruction by point set structuring [EG13] Multi-view stereo reconstruction by multiple shape sampling [CVPR10, PAMI13] Reconstruction of indoor scenes by multi-layered 2D-arrangements [JPRS14] |
|
Surface approximation |
 |
Surface approximation typically consists in remeshing a input surface by imposing constraints on the geometry, topology, complexity or structure of the output solution. I am particularly interested in designing methods that preserve structural considerations.
Main contributions: Mesh decimation driven by planar shapes [CGF15] Constrained remeshing in Zometool models [GM14] |
|
Object and scene classification |
 |
Object and scene classification is a traditional problem in vision and machine learning. I am interested in designing classification methods both fast and scalable by exploiting geometric information at a high level.
Main contributions: Classification of textured meshes for urban landscapes [JPRS17] Classification of man-made objects by planar shape analysis [ISPRS16] Pyramid scene parsing Network for 3D point cloud classification [JPRS19] |
|
Sketch interpretation |
 |
Sketches drawn by designers are traditionally bitmap images that need to be converted into vector graphics for better rendering quality, editability and compactness. I am interested in vectorizing line-drawings and multilayered color images using non-local analysis, as well as interpreting sketches in 3D using geometric priors.
Main contributions: Vectorization of line-drawings by Bezier curve networks [SIGGRAPH16] 3D reconstruction of line-drawings in a multiview stereo context [CVPR15] Image Abstraction by Layered Linear Color Gradients[SIGGRAPH ASIA17] |
|
Student supervision |
| Zhenyu Zhu | Object vectorization in remote sensing data [PhD., 2024-2027] |
| Marion Boyer | Building reconstruction from satellite imagery [PhD., 2022-2025] |
| Mulin Yu | Remeshing BIM models [PhD., 2019-2022] |
| Gaetan Bahl | Low-power neural networks [PhD., 2019-2022] |
| Julien Vuillamy | City reconstruction from multi-sourced data [PhD., 2018-2021, supervised with Pierre Alliez] |
| Muxingzi Li | Indoor modeling [PhD., 2018-2021] |
| Jean-Philippe Bauchet | Urban scene reconstruction [PhD., 2016-2019] |
| Oussama Ennafii | Geometric quality of city models [PhD., 2016-2019, supervised with Clement Mallet] |
| Hao Fang | Scale-space exploration [PhD., 2016-2019] |
| Jean-Dominique Favreau | 3D modeling from sketches [PhD., 2014-2017, supervised with Adrien Bousseau] |
| Dorothy Duan | Semantized elevation maps [PhD., 2013-2016] |
| Sven Oesau | 3D Indoor reconstruction [PhD., 2012-2015, supervised with Pierre Alliez] |
| Yannick Verdie | Urban modeling from point clouds [PhD., 2010-2013] |
| Mulin Yu | pliant surface remeshing [Master, 2019, supervised with Pierre Alliez] |
| Alex Auvolat | Surface reconstruction by point processes [Master, 2018, supervised with Adrien Bousseau] |
| Leihan Chen | Floor maps vectorization [Master, 2017] |
| Hao Fang | Multiscale Primitive detection [Master, 2015] |
| Jean-Dominique Favreau | 3D modeling from sketches [Master, 2014, supervised with Adrien Bousseau] |
| Paul Seron | Primitive-driven mesh approximation [Master, 2011, supervised with Pierre Alliez] |
| Ioan Dragan | Parallelization of Markov Random Field [Master 2010-2011] |
| Parmeet Bhatia | Urban library of 3D unitary elements [Master, 2010] |
| Marouene Amri | Classification of Lidar data [Master, 2010] |
| Yann Bhogal | Survey on roof typology [Master, 2006] |
| Pierre Trontin | Building footprint vectorization [Master, 2005] |
|
Organization of scientific events |
|
|