Research map / Point clouds / Shape and surface completion
Neu-PiG: Neural Preconditioned Grids for Fast Dynamic Surface Reconstruction on Long Sequences
Julian Kaltheuner, Hannah Dröge, Markus Plack, Patrick Stotko, Reinhard Klein
Read the paperCheck your idea against 137,217 papers
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Sobolev preconditioning to latent grid training (ablation) | Neu-PiG · DT4D | Chamfer distance: 0.98 -> 0.87 | Improved |
| use multi-resolution grid instead of hash encoding (ablation) | Neu-PiG · DT4D | Chamfer distance: 1.23 -> 0.87 | Improved |
| use eight-level multi-resolution grid instead of single level (ablation) | Neu-PiG · DT4D | Chamfer distance: 0.98 -> 0.87 | Improved |
| add normal-direction latent encoding zn (ablation) | Neu-PiG · DT4D | Chamfer distance: 0.91 -> 0.87 | Improved |
| add isometry loss (ablation) | Neu-PiG · DT4D | Chamfer distance: 1.03 -> 0.87 | Improved |
| use Fourier time encoding instead of Gaussian Fourier (ablation) | Neu-PiG · AMA | Chamfer distance: 0.77 -> 0.44 | Improved |
| use Fourier time encoding instead of polynomial (ablation) | Neu-PiG · AMA | Chamfer distance: 0.46 -> 0.44 | Improved |
| add catch-up variable delta schedule (ablation) | Neu-PiG · AMA T=40 | Chamfer distance: 3.22 -> 0.53 | Improved |
| use cumulative temporal weighting omega (ablation) | Neu-PiG · AMA T=40 | Chamfer distance: 1.15 -> 0.53 | Improved |
| train 1000 epochs instead of 250 (ablation) | Neu-PiG · DFAUST | Chamfer distance: 0.40 -> 0.39 | No clear effect |
| train 1000 epochs instead of 250 (ablation) | Neu-PiG · AMA | Chamfer distance: 0.53 -> 0.44 | Improved |
| replace per-timestep deformation grids (PDG) with shared latent grid | PDG · AMA T=120 | Chamfer distance: 30.20 -> 1.31 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DynoSurf — DynoSurf: Neural Deformation-based Temporally Consistent Dynamic Surface Reconstruction (ECCV 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Motion2VecSets: 4D Latent Vector Set Diffusion for Non-rigid Shape Reconstruction and Tracking (CVPR 2024 · similarity 0.88)
- 4Deform: Neural Surface Deformation for Robust Shape Interpolation (CVPR 2025 · similarity 0.88)
- TFS-NeRF: Template-Free NeRF for Semantic 3D Reconstruction of Dynamic Scene (NeurIPS 2024 · similarity 0.88)
- SC-NeuS: Consistent Neural Surface Reconstruction from Sparse and Noisy Views (AAAI 2024 · similarity 0.88)
- NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse Input Views (AAAI 2024 · similarity 0.87)
- Implicit Neural Surface Deformation with Explicit Velocity Fields (ICLR 2025 · similarity 0.87)
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.
