Noisy SDF targets generated from a second, independently noisy point cloud can supervise a neural network to predict nearly clean signed distance fields.
Approximating the signed distance with respect to the perturbed reference˜pand treatingnas constant to first order, we obtain ˜s≈n⊤(q−˜p) =n ⊤(q−(p+ε)) =n ⊤(q−p)−n ⊤ε≈s−n ⊤ε
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
NoiseSDF2NoiseSDF: Learning Clean Neural Fields from Noisy Supervision
Noisy SDF targets generated from a second, independently noisy point cloud can supervise a neural network to predict nearly clean signed distance fields.