Research map / Point clouds / Shape and surface completion
GridPull: Towards Scalability in Learning Implicit Representations from 3D Point Clouds
Chao Chen, Yu-Shen Liu, Zhizhong Han
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 gradient-consistency loss to pulling loss (ablation) | GridPull · FAMOUS | CD_L2 x100: 0.184 -> 0.126 | Improved |
| add total variation continuity loss on grid distances (ablation) | GridPull · FAMOUS | CD_L2 x100: 0.126 -> 0.056 | Improved |
| add surface loss on input points (ablation) | GridPull · FAMOUS | CD_L2 x100: 0.056 -> 0.040 | Improved |
| use geometric sphere initialization instead of random (ablation) | GridPull · FAMOUS | CD_L2 x100: 0.155 -> 0.040 | Improved |
| add Eikonal constraint on gradients (ablation) | GridPull · FAMOUS | CD_L2 x100: 0.040 -> 0.054 | Hurt |
| restrict pulling loss to narrow bandwidth near surface (M1=3) instead of all grids (ablation) | GridPull · FAMOUS | time: 628s -> 84s; CD 0.043 -> 0.040 | Improved |
| increase TV bandwidth M2 from 2 to 14 (ablation) | GridPull · FAMOUS | CD_L2 x100: 0.079 -> 0.040 | Improved |
| increase TV bandwidth beyond 14 to all grids (ablation) | GridPull · FAMOUS | CD_L2 x100: 0.040 -> 0.040; time 84s -> 101s | No clear effect |
| increase grid resolution R from 256 to 512 (ablation) | GridPull · FAMOUS | CD_L2 x100 and time: 0.040 -> 0.035; 201s -> 645s | Mixed |
| add ground-truth normal supervision (ablation) | GridPull · FAMOUS | CD_L2: 0.0401 -> 0.0403 | No clear effect |
| replace neural network with discrete grid SDF (ablation) | NeuralPull-style pulling · FAMOUS | CD_L2 x100: NeuralPull+L_grad+L_Surface 0.203 vs GridPull 0.040; time 554s vs 84s | Improved |
| scale number of input points from 10K to 1000K (ablation) | GridPull · FAMOUS | time: 82s -> 90s | No clear effect |
Later papers that compared with this method
2 later accepted papers report GridPull as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- I-Filtering: Implicit Filtering for Learning Neural Distance Functions From 3D Point Clouds (TPAMI 2025 · better on 3 of 3 shared results)
- Implicit Filtering for Learning Neural Signed Distance Functions from 3D Point Clouds (ECCV 2024 · better on 2 of 3 shared results)
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.
- NeuralPull — Neural-Pull: Learning Signed Distance Function from Point clouds by Learning to Pull Space onto Surface (ICML 2021) · better on 3 of 3 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
- MultiPull: Detailing Signed Distance Functions by Pulling Multi-Level Queries at Multi-Step (NeurIPS 2024 · similarity 0.91)
- GridFormer: Point-Grid Transformer for Surface Reconstruction (AAAI 2024 · similarity 0.90)
- GeoUDF: Surface Reconstruction from 3D Point Clouds via Geometry-guided Distance Representation (ICCV 2023 · similarity 0.90)
- A Survey and Benchmark of Automatic Surface Reconstruction From Point Clouds (TPAMI 2024 · similarity 0.90)
- Iso-Points: Optimizing Neural Implicit Surfaces With Hybrid Representations (CVPR 2021 · similarity 0.90)
- Arbitrary-Scale Point Cloud Upsampling by Voxel-Based Network with Latent Geometric-Consistent Learning (AAAI 2024 · similarity 0.88)
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.
