Compositional SDF models often produce objects that interpenetrate. Soft penalties reduce this without eliminating it. In a paper to appear at #NeurIPS, we enforce hard non-intersection constraints plug-and-play and with negligible overhead.
https://t.co/2ksbxsRRbw
#3DVision
Compositional SDF models often produce objects that interpenetrate. Soft penalties reduce this without eliminating it. In a paper to appear at #NeurIPS, we enforce hard non-intersection constraints plug-and-play and with negligible overhead.
https://t.co/2ksbxsRRbw
#3DVision
Gradient-based shape optimization in latent space drifts off the valid-shape manifold.
Our fix to appear at NeurIPs: alternate gradient steps with guided flow matching. Optimize freely, correct strictly.
Tested on drag reduction & compliance.
https://t.co/BYE0vwjAoW
Gradient-based shape optimization in latent space drifts off the valid-shape manifold.
Our fix to appear at NeurIPs: alternate gradient steps with guided flow matching. Optimize freely, correct strictly.
Tested on drag reduction & compliance.
https://t.co/BYE0vwjAoW
We argue that extracting explicit surfaces from neural unsigned distance fields benefits from multi-pass approaches, as they can refine surface predictions over time! Compared to perfect fields, neural ones have noisy spots which are missed by single-pass meshing techniques.
If you are at @NeurIPSConf in San Diego come say hi! I'll be presenting my latest paper "High Resolution UDF Meshing via Iterative Networks" at the poster session on Friday from 11am to 2pm🧑🔬
https://t.co/4uGMc7X4GC
#NeurIPS2025#NeurIPS#Research#AI#ComputerVision
While Unsigned Distance Fields (UDFs), unlike Signed Distance Fields (SDFs), can naturally represent open surfaces, triangulated them is challenging. In a paper to appear in NeurIPs, we propose an effective iterative refinement approach. https://t.co/FJ7UFm07no #ai
Surfaces can easily be extracted from Signed Distance Fields, but not from Unsigned Distance ones. In an upcoming ECCV paper ( https://t.co/FJ7UFm07no), we introduce a deep-learning based approach to locally turning a UDF into an SDF so that traditional algorithms can be used.