Presentation

<
SharpNet: Enhancing MLPs to Represent Functions with Controlled Non-differentiability
SessionDifferentiable Geometry Processing & Flow Maps
DescriptionWe propose SharpNet, an enhanced MLP architecture that augments the input with a feature function to produce controllable derivative jumps. We show that SharpNet can preserve the sharpness in the intended regions while maintaining smoothness elsewhere, and that it outperforms baseline methods in sharpness-critical tasks such as CAD reconstruction.
Authors
Hanting Niu
Institute of Software, Chinese Academy of Sciences
University of Chinese Academy of Sciences
Junkai Deng
Nanyang Technological University
Fei Hou
Institute of Software, Chinese Academy of Sciences
University of Chinese Academy of Sciences
Wencheng Wang
Institute of Software, Chinese Academy of Sciences
University of Chinese Academy of Sciences
Ying He
Nanyang Technological University
Event Type
Technical Paper
TimeWednesday, 22 July 20262:10pm - 2:20pm PDT
LocationRoom 403 B
Digital Library PDF
Session Time & Location
Sunday, 19 July 20266:00pm - 8:45pm PDTHall K
Wednesday, 22 July 20262:00pm - 3:30pm PDTRoom 403 B
Registration Categories
Full Conference Supporter
Full Conference
Next PresentationNext PresentationSchrödinger Bridges on Discretized Geometric Domains
Similar Presentations
SuperSDF:Sparse SDF Super-Resolution for Surface Extraction
MegaNorm: Local Patches Embedding for Efficient and Robust Point Normal Orientation at Super-Large Scale
Dual Contouring over Expanded Cubes (DCx) for Zero-Level Set Extraction from Neural Unsigned Distance Functions