GIC: Gaussian-Informed Continuum for Physical Property Identification and Simulation
Authors: Junhao Cai, Yuji Yang, Weihao Yuan, Yisheng HE, Zilong Dong, Liefeng Bo, Hui Cheng, Qifeng Chen
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental evaluations demonstrate that our pipeline achieves state-of-the-art performance across multiple benchmarks and metrics. Additionally, we illustrate the effectiveness of the proposed method through real-world demonstrations, showcasing its practical utility. |
| Researcher Affiliation | Collaboration | 1The Hong Kong University of Science and Technology, 2Sun Yat-sen University, 3Alibaba Group |
| Pseudocode | Yes | The proposed strategy is presented in Alg. 1. |
| Open Source Code | No | The code is not included for now. But we will release the code to the public soon. |
| Open Datasets | Yes | To thoroughly assess our proposed method, we employ two sources of data introduced by PAC-Ne RF [12] and Spring-Gaus [6]. |
| Dataset Splits | No | We follow the setting in Spring-Gaus [6] that uses the first 20 frames as training data and the subsequent 10 frames for evaluation. |
| Hardware Specification | Yes | All the experiments are conducted on a single A10 GPU. |
| Software Dependencies | No | The paper mentions using Adam optimizer and refers to settings from other papers, but does not provide specific version numbers for software dependencies or libraries. |
| Experiment Setup | Yes | We employ the same setting in [16] to train our pipeline. Concretely, the canonical 3D Gaussians are initialized with points randomly sampled from the given bounding box of the scene. We start training the deformation network after 3,000 iterations of warm-up for the 3D Gaussians. The total number of iterations is set at 40,000, with densification and pruning operations conducted every 500 steps until reaching 15,000 iterations. Additionally, the number of motions Nm is set to 8 for all objects in our network. λ1 and λ2 in Eqn. 4 are all set to 1. In Alg. 1, the number of iterations, denoted as nu, is uniformly set to 4 for all objects. We set the initial grid size x according to the volume of the object. For most objects, x = 0.1, while for small items such as toothpaste in PAC-Ne RF dataset, x = 0.01. The parameters thmin and thmax are set to 0.5 and 0.8, respectively. |