SNAKE: Shape-aware Neural 3D Keypoint Field
Authors: Chengliang Zhong, Peixing You, Xiaoxue Chen, Hao Zhao, Fuchun Sun, Guyue Zhou, Xiaodong Mu, Chuang Gan, Wenbing Huang
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We achieve superior performance on various public benchmarks, including standalone object datasets Model Net40, Keypoint Net, SMPL meshes and scene-level datasets 3DMatch and Redwood. Via extensive quantitative and qualitative evaluations on standalone object datasets Model Net40, Keypoint Net, SMPL meshes, and scene-level datasets 3DMatch and Redwood, we demonstrate that our shape-aware formulation achieves state-of-the-art performance under three settings: (1) semantic consistency; (2) repeatability; (3) geometric registration. |
| Researcher Affiliation | Collaboration | 1Xi an Research Institute of High-Tech 2THUAI, Tsinghua University 3AIR, Tsinghua University 4 Peking University 5Intel Labs 6MIT 7Gaoling School of Artificial Intelligence, Renmin University of China 8Beijing Key Laboratory of Big Data Management and Analysis Methods |
| Pseudocode | Yes | Algorithm 1 Optimization for Explicit Keypoint Extraction |
| Open Source Code | Yes | Codes are available at https://github.com/zhongcl-thu/SNAKE. |
| Open Datasets | Yes | The Keypoint Net [39] dataset and meshes generated with the SMPL model [18] are utilized. Model Net40 [37] is a synthetic object-level dataset. 3DMatch [41] and Redwood [5] are RGB-D reconstruction datasets. |
| Dataset Splits | Yes | We adopt the official dataset split strategy. train the model on 3DMatch and test it on Redwood. We randomly perform SE(3) transformation on the test point clouds to generate the second view point clouds. |
| Hardware Specification | No | The paper mentions “computational cost and time” and refers to a “comparison of computation cost” in the Appendix, but it does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. |
| Experiment Setup | Yes | The implementation details and hyper-parameters for SNAKE in three settings can be found in the Appendix. The learning rate for optimization is also key to the final result. When the learning rate is set to 0.1, 0.01, 0.001 and 0.0001, the relative repeatability (%) on Model Net40 dataset with the same experimental settings as Table 6 are 0.002, 0.622, 0.854 and 0.826, respectively. |