Résumé
In the realm of modern digital phenotyping technologicaladvancements, the demand of annotated datasets is increasing for eithertraining machine learning algorithms or evaluating 3D phenotypingsystems. While a few 2D datasets have been proposed in the communityin last few years, very little attention has been paid to the constructionof annotated 3D point cloud datasets. There are several challengesassociated with the creation of such annotated datasets. Acquiringthe data requires instruments having good precision and accuracylevels. Reconstruction of full 3D model from multiple views is a challengingtask considering plant architecture complexity and plasticity, aswell as occlusion and missing data problems. In addition, manual annotationof the data is a cumbersome task that cannot easily be automated.In this context, the design of synthetic datasets can play animportant role. In this paper, we propose an idea of automatic generationof synthetic point cloud data using virtual plant models. Our approachleverages the strength of the classical procedural approach (likeL-systems) to generate the virtual models of plants, and then performpoint sampling on the surface of the models. By applying stochasticityin the procedural model, we are able to generate large number of diverseplant models and the corresponding point cloud data in a fully automaticmanner. The goal of this paper is to present a general strategyto generate annotated 3D point cloud datasets from virtual models. Thecode (along with some generated point cloud models) are available at:https://gitlab.inria.fr/mosaic/publications/lpy2pc.