TopoGaussian: Inferring Internal Topology Structures from Visual Clues
Authors: Xiaoyu Xiong, Changyu Hu, Chunru Lin, Pingchuan Ma, Chuang Gan, Tao Du
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the efficacy of our pipeline on a synthetic dataset and four real-world tasks with 3D-printed prototypes. Compared with existing mesh-based method, our pipeline is 5.26x faster on average with improved shape quality. We present a synthetic dataset and perform benchmark experiments comparing our method with two strong mesh-based baselines PGSR (Chen et al., 2024) and Gaussian Surfels (Dai et al., 2024), based on the processing time and reconstruction quality (3D printability). Moreover, we extend our method to more synthetic validations, four real world validations, and several ablation studies, which show the ability of our pipeline to handle input under various conditions. |
| Researcher Affiliation | Academia | Xiaoyu Xiong1, Changyu Hu1, Chunru Lin2, Pingchuan Ma3, Chuang Gan2, Tao Du14 1 Tsinghua University 2 University of Massachusetts Amherst 3 Massachusetts Institute of Technology 4 Shanghai Qi Zhi Institute |
| Pseudocode | No | The paper includes a section "A DETAILS OF ALGORITHMS" which contains mathematical derivations and descriptions of methods, but it does not present any structured pseudocode or algorithm blocks with numbered steps formatted like code. |
| Open Source Code | No | The paper includes a footnote "1https://topo-gaussian.github.io/Topo Gaussian/" on the first page, which is a project page. However, it does not explicitly state that source code is released or provide a direct link to a code repository. |
| Open Datasets | No | The paper mentions using a "synthetic dataset" and conducting "real-world tasks" but does not provide concrete access information such as a URL, DOI, repository name, or formal citation for any publicly available or open dataset used in the experiments. |
| Dataset Splits | No | The paper discusses the use of a synthetic dataset and neural implicit surface training but does not provide specific details regarding dataset splits, such as exact percentages, sample counts for training, validation, or test sets, or references to predefined splits. |
| Hardware Specification | Yes | We evaluated our pipeline on the tasks below on a server with an AMD EPYC 9754 128-Core CPU, 12 DDR5 4800 16GB (384GB in total) RAM, and 1 NVIDIA RTX 4090 24G GPU. |
| Software Dependencies | No | The paper mentions several software components like C++, Python, CUDA, Open MP, Nvidia Warp, Open3D, Open CV, Py Torch, L-BFGS-B, and Adam. However, it does not provide specific version numbers for most of these key libraries or frameworks, which is necessary for a reproducible description of ancillary software. |
| Experiment Setup | Yes | Neural Implicit Surface: We train for 10000 epochs with learning rate 3e-6. Simulation: The basic simulation parameters are listed in Tab. 2, while the actuation pressure of the soft hand is set to be 5e5 Pa. Table 2 provides specific values for 'Time step (s)', 'Density (kg/m3)', 'Young s modulus (Pa)', and 'Poisson s ratio' for various experiment types (Rigid, Soft, Rigid (real), Soft (real)). |