Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
MIND: Material Interface Generation from UDFs for Non-Manifold Surface Reconstruction
Authors: Xuhui Chen, Fei Hou, Wencheng Wang, Hong Qin, Ying He
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on UDFs generated from diverse data sources, including point cloud reconstruction, multi-view reconstruction, and medial axis transforms, demonstrate that our approach robustly handles complex non-manifold surfaces and significantly outperforms existing methods. |
| Researcher Affiliation | Academia | 1Key Laboratory of System Software (CAS) and State Key Laboratory of Computer Science, Institute of Software, Chinese Academy of Sciences 2University of Chinese Academy of Sciences 3Department of Computer Science, Stony Brook University 4College of Computing and Data Science, Nanyang Technological University |
| Pseudocode | No | The paper describes its methodology in Section 3 ("Methodology") and its subsections (3.1, 3.2, 3.3) using textual descriptions and mathematical equations. No explicit 'Pseudocode' or 'Algorithm' blocks are present. |
| Open Source Code | Yes | The source code is available at https://github.com/jjjkkyz/MIND. |
| Open Datasets | Yes | Table 1: Evaluation on the Shape Net-Car [62] dataset and Deep Fashion3D dataset [63]. |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits for the datasets used in its evaluation (Shape Net-Car, Deep Fashion3D), nor for the UDF learning methods it builds upon. It mentions evaluating on these datasets without specifying how they were partitioned for evaluation. |
| Hardware Specification | Yes | All results are tested on a single NVIDIA V100 GPU. |
| Software Dependencies | No | We use M3C [27] to extract meshes, implemented in Dream3D4. (Footnote 4 links to Dream3D GitHub but no specific version number is mentioned for Dream3D or M3C.) |
| Experiment Setup | Yes | We normalize 3D models to fit within [ 0.5, 0.5]3 and use a bounding box of [ 0.6, 0.6]3 to contain the UDFs. For calculating the local two-signed field, we sample 1 million points on the r1 level set and optimize their positions to align with the local minima of the UDF. The resolution is set to 2563, resulting in a voxel size of 0.0046. The voxel size for point cloud downsampling3 is set to 0.005... we set r2 = 0.01... r1 is set to 0.05. We erode the local two-signed field 2 times... We then optimize the output of M3C with Equation 4 for 200 iterations, using a Laplacian weight of 1000, to generate the final result. |