SGI Projects
MIT's Summer Geometry Initiative is an intensive introduction to concepts in the field of geometry processing. I had the opportunity to participate as a fellow in 2022, during which time I worked with researchers on three projects.
(Download Slides)Orienting Point Clouds
Michael Kazhdan - JHU, Tiago De Souza Fernandes, Anna Krokhine, Qi Zhang
Blog Post
- Deals with the classic problem of assigning normals to unoriented points in a set — older 3D scans may lack normal information
-
Two-phase approach
- Local phase: points are clustered according to their k-nearest neighbors and oriented via fitting to an implicit surface representation
-
Global phase: assign signs to each clusters in a greedy fashion
- Try all 2c combinations of clusters and see which one minimizes an alignment energy
- Merge one level up, then repeat until convergence
- Alignment energy of neighboring points: based on the average directional similarity when projected onto the other point's corresponding implicit surface
- We turn this into an energy based on signed clusters as follows:
- Development of new implicit surface representations
- Parabolic Fit (dual-axis parabola)
- "Associative" Fit (hybrid approach that compares two surface types, e.g. plane & unconstrained quadratic)
- Wedge Fit (preserves detail at sharp corners)
