Normal Adaptive Neural Meshes
Ioannis Ivrissimtzis,
Won-Ki Jeong, and Hans-Peter Seidel.
The Neural Mesh learns an unorganized point cloud through a
competitive Learning process. The Network expands incrementally by
duplicating the nodes with the greatest role in the representation of
the point cloud, and pruning the least important ones. Statistically
based operators create boundaries and handles, giving the Neural Mesh
the capability to learn topologies. In the figure the algorithm is
modified to reconstruct a curvature adaptive mesh directly from the
point cloud. This is achieved by studying the variation of the Neural
Mesh normals during the learning process.
Back to AG4 Gallery
