Research map / Point clouds / Shape and surface completion
Differentiable Spline Approximations
Minsu Cho, Aditya Balu, Ameya Joshi, Anjana Deva Prasad, Biswajit Khara, Soumik Sarkar, Baskar Ganapathysubramanian, Adarsh Krishnamurthy, Chinmay Hegde
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 a piecewise constant DSA layer | UNet segmentation model · Weizmann Horse | Jaccard score: 72.06 ± 0.60 -> 73.13 ± 0.31 | Improved |
| Add a piecewise constant DSA layer | UNet segmentation model · Broad Bioimage Benchmark | Jaccard score: 79.34 ± 0.43 -> 81.56 ± 0.24 | Improved |
| Allow knot locations to change | DSA NURBS surface fitting · Bukin function N.6, 128×128 points | mean squared error: 19.83 ± 0.001 -> 8.25 ± 0.01 | Improved |
| Allow knot locations to change | DSA NURBS surface fitting · Bukin function N.6, 256×256 points | mean squared error: 19.85 ± 0.001 -> 8.23 ± 0.02 | Improved |
| Use DSA surface evaluation and unsupervised training without control-point regression loss | SplineNet · Spline Dataset test set | Chamfer distance × 100: 1.18 ± 0.10 -> 0.03 ± 0.02 | Improved |
| Use 5×5 rather than 20×20 control points (ablation) | DSA SplineNet reconstruction · Spline Dataset test set | Chamfer distance × 100: 0.03 ± 0.02 -> 0.14 ± 0.07 | Hurt |
| Use 4×4 rather than 20×20 control points (ablation) | DSA SplineNet reconstruction · Spline Dataset test set | Chamfer distance × 100: 0.03 ± 0.02 -> 0.02 ± 0.01 | Improved |
| Use DSA with degree-1 Lagrange polynomials instead of PINN | Poisson equation solver · Poisson equation, 128×128 | L2 norm: 3.72 ± 0.20 E-4 -> 3.32 ± 0.05 E-5 | Improved |
| Use DSA with degree-1 Lagrange polynomials instead of PINN | Poisson equation solver · Poisson equation, 256×256 | L2 norm: 2.63 ± 0.20 E-4 -> 2.57 ± 0.01 E-5 | Improved |
| Use degree-2 rather than degree-1 Lagrange polynomials (ablation) | DSA Poisson equation solver · Poisson equation, 128×128 | L2 norm: 3.32 ± 0.05 E-5 -> 2.16 ± 0.04 E-5 | Improved |
| Increase polynomial degree from 1 to 2 or 3 (ablation) | DSA Poisson equation solver · Poisson equation, 256×256 | L2 norm: 2.57 ± 0.01 E-5 -> 2.79 ± 0.20 E-5; 2.57 ± 0.01 E-5 -> 2.59 ± 0.10 E-5 | Mixed |
Closest papers by meaning
- A Fully First-Order Layer for Differentiable Optimization (ICML 2026 · similarity 0.86)
- Differentiable Convex Optimization Layers (NeurIPS 2019 · similarity 0.86)
- The Multi-Block DC Function Class: Theory, Algorithms, and Applications (NeurIPS 2026 · similarity 0.84)
- Optimizing $(L_0, L_1)$-Smooth Functions by Gradient Methods (ICLR 2025 · similarity 0.84)
- Directional Smoothness and Gradient Methods: Convergence and Adaptivity (NeurIPS 2024 · similarity 0.84)
- Nonlinearly Preconditioned Gradient Methods: Momentum and Stochastic Analysis (NeurIPS 2025 · similarity 0.83)
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.
