AI Paper Atlas

Research map / Point clouds / Shape and surface completion

Towards Better Gradient Consistency for Neural Signed Distance Functions via Level Set Alignment

Baorui Ma, Junsheng Zhou, Yu-Shen Liu, Zhizhong Han

CVPR 2023 · Point clouds › Shape and surface completion · approach: Distance-function surface reconstruction · 41 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add level set alignment loss to NeuralPullNeuralPull · 3D SceneCD: 0.059 -> 0.020Improved
add level set alignment loss to NeuSNeuS · DTUChamfer distance: 0.87 -> 0.78Improved
add level set alignment loss to MonoSDFMonoSDF · ScanNetF-score: 0.733 -> 0.750Improved
increase alpha from 0 to 0.01, then increase it to 1.0 (ablation)NeuralPull with level set alignment · 3D SceneCD×100: 0.586 -> 0.394 -> 0.601Mixed
add adaptive per-query weighting rather than using no weight (ablation)NeuralPull with level set alignment · 3D SceneCD×100: 0.455 -> 0.394Improved
use inferred signed distance for adaptive weights instead of point-cloud Euclidean distance (ablation)NeuralPull with level set alignment · 3D SceneCD×100: 0.394 vs 0.391No clear effect
use decay parameter delta 10 rather than 1 or 100 (ablation)NeuralPull with level set alignment · 3D SceneCD×100: 0.439 -> 0.394; 0.394 -> 0.489Mixed
use current zero-level-set projections instead of fixed nearest surface points (ablation)NeuralPull with level set alignment · 3D SceneCD×100: 1.384 -> 0.394Improved
use cosine distance instead of MSE on normalized gradients (ablation)NeuralPull with level set alignment · 3D SceneCD×100: 0.486 -> 0.394Improved
use cosine distance instead of MSE on unnormalized gradients (ablation)NeuralPull with level set alignment · 3D SceneCD×100: 0.494 -> 0.394Improved
use cosine level-set alignment instead of the DiGS second-order constraint (ablation)NeuralPull · 3D SceneCD×100: 0.601 -> 0.394Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy