Machines might have a sense for Christmas after all!
This scene of three trees was reconstructed from an unori-ented point cloud with scan shadows. Our neural network Points2Surf [1] estimates a signed-distance function from a local and a global subset of the point cloud. By relying not only on global data, it is independent from object classes.

Scan the QR code or follow the link on the right to visit our procedural Christmas song generator. It uses a musi-cal grammar [2] inspired by split and shape grammars. Feel free to play around with the exposed parameters or let the random generator surprise you.

[1] Erler, Philipp et al. “Point2Surf: Learning Implicit Surfaces from Point Clouds” in European Conference on Computer Vision
[2] Eibensteiner, Lukas et at. “Temporal-Scope Grammars for Polyphonic Music Generation” in ACM SIGPLAN Int. Workshop on Func. Art, Music, Modeling, and Design