KT-Net: Knowledge Transfer for Unpaired 3D Shape Completion
Authors: Zhen Cao, Wenxiao Zhang, Xin Wen, Zhen Dong, Yu-Shen Liu, Xiongwu Xiao, Bisheng Yang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct comprehensive experiments on several datasets, and the results show that our method outperforms previous methods of unpaired point cloud completion by a large margin. |
| Researcher Affiliation | Collaboration | 1Wuhan University 2Singapore University of Technology and Design 3JD Logistics 4Tsinghua University |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/a4152684/KT-Net. |
| Open Datasets | Yes | We evaluate our network on two systhetic datasets 3D-EPN (Dai, Ruizhongtai Qi, and Nießner 2017) and CRN (Wang, Ang Jr, and Lee 2020), and three real-world datasets Matter Port3D (Chang et al. 2017), Scan Net (Dai et al. 2017) and KITTI (Geiger, Lenz, and Urtasun 2012). |
| Dataset Splits | Yes | For evaluation on synthetic datasets, we use Chamfer Distance(CD) to compare the difference between the predicted point cloud and the ground truth. For the real-world datasets without ground truth, we follow Shape Inversion(Zhang et al. 2021) to use Unidirectional Hausdorff Distance(UHD) to evaluate the similarity between the input point cloud and the predicted point cloud. ... In this section, we follow the train/test split of the previous unpaired shape completion methods (i.e., Pcl2Pcl and Cycle4Completion) for a fair comparision. |
| Hardware Specification | No | The paper does not provide specific details regarding the hardware specifications (e.g., GPU models, CPU types, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., libraries, frameworks, or specific tools). |
| Experiment Setup | Yes | For a given learning rate γ, we first update θD of all the discriminators and freeze all other parts... where λg and λp are hyper-parameters and we set λg = 0.1 and λp = 1. |