NeuralSlice: Neural 3D Triangle Mesh Reconstruction via Slicing 4D Tetrahedral Meshes
Authors: Chenbo Jiang, Jie Yang, Shwai He, Yu-Kun Lai, Lin Gao
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate that our method seeks a new solution to describing the diverse topology in an explicit fashion and outperforms existing explicit approaches with high accuracy and robustness for 3D shape reconstruction. Our method is superior to implicit approaches with 10 times faster speed for shape reconstruction and improved robustness to thin structures. To summarize, our major contributions are three-fold: [...] Experiments demonstrate that our 3D reconstruction method accurately reconstructs 3D shapes of diverse topology, outperforming existing explicit methods in accuracy, and much faster than implicit methods. Furthermore, NEURALSLICE can represent different 3D shapes and topologies in one 4D tetrahedral mesh. [...] After describing the experiment settings, we first evaluate our method on reconstructing 3D meshes from point clouds, and compare it to state-of-the-art methods. We then conduct ablation studies to verify the effectiveness of each design choice. Finally, we analyze the limitations and failure cases of our method. |
| Researcher Affiliation | Academia | 1Beijing Key Laboratory of Mobile Computing and Pervasive Device, Institute of Computing Technology, Chinese Academy of Sciences 2Nanjing University of Science and Technology Zijin College 3University of Maryland, College Park 4Cardiff University 5University of Chinese Academy of Sciences. |
| Pseudocode | Yes | Details and pseudocode are in Appendix B.2 and B.3. [...] def slice(vertices, alpha): |
| Open Source Code | Yes | The corresponding code can be found on Git Hub at https: //github.com/IGLICT/NEURALSLICE. |
| Open Datasets | Yes | We use the Shape Net (Chang et al., 2015) dataset for training and evaluation, utilizing the 13 categories and the same dataset split as in (Gupta & Chandraker, 2020). |
| Dataset Splits | Yes | We use the Shape Net (Chang et al., 2015) dataset for training and evaluation, utilizing the 13 categories and the same dataset split as in (Gupta & Chandraker, 2020). |
| Hardware Specification | Yes | Our experiments were carried out on a computer with an Intel 10700 CPU and an RTX 3090 GPU. |
| Software Dependencies | No | Our method is implemented using Py Torch (Paszke et al., 2019). We use the Adam optimizer (Kingma & Ba, 2015). No specific version numbers for PyTorch or the Adam optimizer are provided. |
| Experiment Setup | Yes | We use the Adam optimizer (Kingma & Ba, 2015) to train our NEURALSLICE without the refinement net for 150 epochs with batch size of 32 and learning rate of 10 3. We then train NEURALSLICE with the refinement net for 150 epochs with batch size of 16 and learning rate of 10 4. For hyper-parameters, we set λlap = 0.1, Li = 1.12 ( i) empirically. |