There has been recently a growing interest for implicit shape. Contrary to explicit representations, they have no resolution and they easily deal with a wide variety of surface topologies. To these implicit representations, current approaches rely on a certain of shape supervision (e.g., inside/outside information or-to-shape knowledge), or at least require a dense point cloud (to well enough the distance-to-shape). In contrast, we introduce, a self-supervised method for learning shape representations from extremely sparse point clouds. Like in Buffon's needle problem, we"drop" (sample) needles on the point cloud and consider that, statistically, to the surface, the needle end points lie on opposite sides of the. No shape knowledge is required and the point cloud can be highly, e.g., as lidar point clouds acquired by vehicles. Previous-supervised shape representation approaches fail to produce good-quality on this kind of data. We obtain quantitative results on par with supervised approaches on shape reconstruction datasets and show qualitative results on hard autonomous driving datasets such as.
No takes yet. Share an insight, caveat, or question.
Boulch et al. (2021) studied this question.