Parameterization Invariant Representations for Efficient Shape Learning

Emmanuel Hartman (U of Houston)

Feb 10. 2025, 15:30 — 16:20

We present a data-driven framework for parameterization invariant representations of 3d graphs and meshes based on the gradient of the varifold norm. In addition to parameterization invariance, these representations are robust to sampling noise have a fixed dimension regardless the number of vertices of the input. As such the proposed representations can be used with traditional neural network architectures for downstream learning tasks such as classification and registration of data from raw scans.

 

Further Information
Venue:
ESI Boltzmann Lecture Hall
Recordings:
Recording
Associated Event:
Infinite-dimensional Geometry: Theory and Applications (Thematic Programme)
Organizer(s):
Tomasz Goliński (U of Białystok)
Gabriel Larotonda (U of Buenos Aires)
Alice Barbara Tumpach (WPI, Vienna)
Cornelia Vizman (WU of Timisoara)