Deep Implicit Surface Point Prediction Networks

Venkatesh, R.; Karmali, Tejan; Sharma, Shruti; Ghosh, Aurobrata; Babu, R. Venkatesh; Jeni, László A.; Singh, Maneesh

doi:10.48550/arxiv.2106.05779

Search citation statements

Order By: Relevance

Paper Sections

Select...

Implicit Function Learning1

Introduction1

Citation Types

Supporting

Mentioning

Contrasting

Year Published

2022

2024

Publication Types

Select...

Other1

Article1

Relationship

Self Cite0

Independent2

Authors

Journals

Cited by 2 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However they have several major limitations: 1) learning UDF is a regression problem, harder than that in SDF; 2) ball pivoting [3] is more computational expensive and less stable than Marching Cubes [25]; 3) gradient vanishes on the surface, resulting in artifacts. Venkatesh et al [39] proposed Closest Surface-Point (CSP) representation to prevent gradient vanishing and improve the speed. Zhao et al [45] proposed Anchor UDF to improve reconstruction accuracy.…”

Section: Implicit Function Learningmentioning

confidence: 99%

See 1 more Smart Citation

3PSDF: Three-Pole Signed Distance Function for Learning Surfaces with Arbitrary Topologies

Chen¹,

Chen²,

Li³

et al. 2022

Preprint

View full text Add to dashboard Cite

Figure 1. We show three groups of shape reconstruction results generated by NDF [10] (in cyan) and our proposed 3PSDF (in gold) respectively. Our method is able to faithfully reconstruct high-fidelity, intricate geometric details including both the closed and open surfaces, while NDF suffers from the meshing problems. Each NDF result is reconstructed from a dense point cloud containing 1 million points while ours are reconstructed using an equivalent resolution.

show abstract

Section: Implicit Function Learningmentioning

confidence: 99%

“…However, they can only model closed surfaces that support the in/out test for level surface extraction. Recent advances that leverage unsigned distance function (UDF) [10,39,40] have made it possible to learn open surfaces from point clouds. But instantiating this field into an explicit mesh remains cumbersome and is prone to artifacts.…”

Section: Introductionmentioning

confidence: 99%