SMA-Net: Deep learning-based identification and fitting of CAD models from point clouds
Article dans une revue avec comité de lecture
Auteur
PERNOT, Jean-Philippe
527033 Laboratoire d'Informatique et des Systèmes (LIS) (Marseille, Toulon) [LIS]
543315 Laboratoire d’Ingénierie des Systèmes Physiques et Numériques [LISPEN]
58355 École Nationale Supérieure des Arts et Métiers [ENSAM]
303092 Arts et Métiers Paristech ENSAM Aix-en-Provence
461986 Institut de recherches économiques et sociales [IRES]
527033 Laboratoire d'Informatique et des Systèmes (LIS) (Marseille, Toulon) [LIS]
543315 Laboratoire d’Ingénierie des Systèmes Physiques et Numériques [LISPEN]
58355 École Nationale Supérieure des Arts et Métiers [ENSAM]
303092 Arts et Métiers Paristech ENSAM Aix-en-Provence
461986 Institut de recherches économiques et sociales [IRES]
Date
2022-04-13Journal
Engineering with ComputersRésumé
Identifcation and ftting is an important task in reverse engineering and virtual/augmented reality. Compared to the traditional
approaches, carrying out such tasks with a deep learning-based method have much room to exploit. This paper presents
SMA-Net (Spatial Merge Attention Network), a novel deep learning-based end-to-end bottom-up architecture, specifcally
focused on fast identifcation and ftting of CAD models from point clouds. The network is composed of three parts whose
strengths are clearly highlighted: voxel-based multi-resolution feature extractor, spatial merge attention mechanism and
multi-task head. It is trained with both virtually-generated point clouds and as-scanned ones created from multiple instances
of CAD models, themselves obtained with randomly generated parameter values. Using this data generation pipeline, the
proposed approach is validated on two diferent data sets that have been made publicly available: robot data set for Industry
4.0 applications, and furniture data set for virtual/augmented reality. Experiments show that this reconstruction strategy
achieves compelling and accurate results in a very high speed, and that it is very robust on real data obtained for instance
by laser scanner and Kinect.
Fichier(s) constituant cette publication
Cette publication figure dans le(s) laboratoire(s) suivant(s)
Documents liés
Visualiser des documents liés par titre, auteur, créateur et sujet.
-
Article dans une revue avec comité de lecturePEUZIN-JUBERT, Manon; NOZAIS, Dominique; MARI, Jean-Luc; PERNOT, Jean-Philippe; POLETTE, Arnaud (Elsevier BV, 2021-12)At present, optical sensors are being widely used to realize high quality control or reverse engineering of products, systems, buildings, environments or human bodies. Although the intrinsic characteristics of such ...
-
Article dans une revue avec comité de lectureSHAH GHAZANFAR, Ali; POLETTE, Arnaud; PERNOT, Jean-Philippe; GIANNINI, Franca; MONTI, Marina (SPRINGER, 2022-03-17)Due to its capacity to evolve in a large solution space, the Simulated Annealing (SA) algorithm has shown very promising results for the Reverse Engineering of editable CAD geometries including parametric 2D sketches, ...
-
Article dans une revue avec comité de lectureSHAH GHAZANFAR, Ali; GIANNINI, Franca; MONTI, Marina; PERNOT, Jean-Philippe; POLETTE, Arnaud (ASME, 2021-12-16)This paper introduces a novel reverse engineering (RE) technique for the reconstruction of editable computer-aided design (CAD) models of mechanical parts’ assemblies. The input is a point cloud of a mechanical parts’ ...
-
Article dans une revue avec comité de lectureMONTLAHUC, Jérémy; SHAH GHAZANFAR, Ali; PERNOT, Jean-Philippe; POLETTE, Arnaud (CAD Solutions LLC (imprimé) and Taylor & Francis Online (en ligne), 2019)This paper introduces a new approach for the generation of as-scanned point clouds of CAD assembly models. The resulting point clouds incorporate various realistic artifacts that would appear if the corresponding real ...
-
Article dans une revue avec comité de lectureThis paper introduces an approach to synthetize new CAD assemblies from existing STEP files. The algorithm first generates linkage graph by detecting linkage between components. Then it detects linkages similarities ...