<p><strong>Context</strong></p>
<p>In a world facing profound upheavals&mdash;ecological, energy-related, economic, health-related, social, and more&mdash;territories are at the heart of the most complex decisions. The stakeholders within these territories are under increasing pressure to anticipate, adapt, and invent new approaches to planning and development. It is now essential to be able to anticipate territorial evolution and simulate different management scenarios in order to assess, and even compare, their impacts. This is the objective of the Digital Twin of France and its Territories (JNFT) project, initiated and co-led by the IGN (National Institute of Geographic and Forest Information), Cerema (Center for Studies and Expertise on Risks, Environment, Mobility and Urban Planning), and Inria (National Institute for Research in Digital Science and Technology).</p>
<p>The 3D reconstruction of urban objects &ndash; in particular the buildings - from remote sensing data - typically Airborne Lidar Scans (ALS) - constitutes one of the core research axes of the JNFT project. Urban reconstruction from Lidar point cloud is a multifaceted problem in which output models are expected to be (i) accurate (adherence to the input data) and (ii) of high geometric quality (conciseness, conforming to urban-specific formalisms and geometric guarantees), and algorithms that produce these models must be (iii) automatic, (iv) fast and scalable, (v) generic to adapt to the variety of urban landscapes, and (vi) geometrically memory-efficient for easy updates of the reconstructed objects along time. The existing methods [1,2,3] offer efficient solutions to criteria (i-iv), but tend to overlook criteria (v) and (vi). In particular, these methods rely upon the construction of space partitioning data structures in 2D and/or 3D that take the form of line and plane arrangements [4,5,6,7]. These data structures allow a limited piecewise-planar description of buildings only, without the capacity of finely approximating non planar components or approaching them by more complex parametric shapes. Once constructed, these data structures can also not be easily modified.</p>
<p>&nbsp;</p>
<p><strong>Objectives</strong></p>
<p>The goal of this position is to investigate a new generation of space partitioning data structures that address the limitations mentioned previously. The candidate will design them in the perspective of replacing the traditional data structures used in the JNFT project, working in collaboration with Luxcarta and GeometryFactory partners and using the CGAL library [8]. The candidate will work on two complementary research directions.</p>
<p>The candidate will investigate how to generalize line and plane arrangements to more expressive data structures that can better capture the geometry of non-planar shapes, e.g. curved facades and freeform roofs. The hybridization of line and plane arrangements with fine mesh-based structures, as proposed in [9], or the use of parametric shapes with a CSG-based construction tree are two options to explore. This generalization will have to be efficient. In particular, it should not degrade the performance in terms of construction time and scalability of the current line and plane arrangements. &nbsp;&nbsp;</p>
<p>The candidate will also investigate the design of update-friendly data structures that can be easily modified once constructed. This characteristic is crucial for efficiently updating the 3D city models, i.e. by only locally modifying the geometry of objects where changes have been detected without reconstructing the entire objects and scenes. Binary Space Partitioning trees, as used in [5,6], could be a solution to this challenge if coupled with an efficient parsing strategy of the geometric atomic elements forming the data structures (e.g. planes and lines).</p>
<p>&nbsp;</p>
<p><strong>Keywords</strong></p>
<p>Geometry processing, Computational geometry, 3D computer vision, geometric data structures, urban reconstruction</p>
<p>&nbsp;</p>
<p><strong>References</strong></p>
<p><strong>[1] </strong>Bauchet, Sulzer, Lafarge and Tarabalka. SimpliCity: Reconstructing Buildings with Simple Regularized 3D Models. CVPRW, 2024</p>
<p><strong>[2] </strong>Peters, Dukai, Vitalis, Van Liempt and Stoter. Automated 3D Reconstruction of LoD2 and LoD1 Models for All 10 million Buildings of the Netherlands. PE&amp;RS journal 2022.</p>
<p><strong>[3] </strong>Huang, Stoter, Peters, and Nan. City3d: Large-scale building reconstruction from airborne lidar point clouds. Remote Sensing, 2022.</p>
<p><strong>[4]</strong> Bauchet and Lafarge. Kinetic Shape Reconstruction. Trans. On Graphics, 2020.</p>
<p><strong>[5]</strong> Sulzer and Lafarge. Concise Plane Arrangements for Low-Poly Surface and Volume Modelling. ECCV 2024</p>
<p><strong>[6] </strong>Pan, Zhang, Liu, Gong and Huang. Building LOD Representation for 3D Urban Scenes. P&amp;RS journal, 2025</p>
<p><strong>[7]</strong> Wu et al. MorphCut: An Efficient Convex Decomposition Method of 3D Building Models for Urban Morphological Analytics. International Journal of Geographical Information Science, 2025</p>
<p><strong>[8]</strong> The CGAL Project. <em>CGAL User and Reference Manual</em>. CGAL Editorial Board, 5.5.1 edition, 2022.</p>
<p><strong>[9]</strong> Jiang et al. Structure-Aware Surface Reconstruction via Primitive Assembly. ICCV 2023.</p>
<p>&nbsp;</p>
<p>More info can be found at&nbsp;<a href="https://team.inria.fr/titane/files/2026/03/sujet_JNFT_geometric_data_structures.pdf">https://team.inria.fr/titane/files/2026/03/sujet_JNFT_geometric_data_structures.pdf</a></p>
<p>&nbsp;</p>
<p>&nbsp;</p>