PaRot: Patch-Wise Rotation-Invariant Network via Feature Disentanglement and Pose Restoration
Authors: Dingxin Zhang, Jianhui Yu, Chaoyi Zhang, Weidong Cai
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that our disentanglement module extracts high-quality rotation-robust features and the proposed lightweight model achieves competitive results in rotated 3D object classification and part segmentation tasks. |
| Researcher Affiliation | Academia | School of Computer Science, University of Sydney, Australia |
| Pseudocode | No | The paper describes the methodology using text and diagrams but does not provide formal pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | We test the classification ability of our model on the synthetic dataset Model Net40 (Wu et al. 2015) and the realworld dataset Scan Object NN (Uy et al. 2019). For shape part segmentation task, we validate our model on the Shape Net Part dataset (Yi et al. 2016). |
| Dataset Splits | No | For Model Net40, the dataset is split into 9,843 training samples and 2,468 testing samples. For Shape Net Part dataset, it contains 14,006 training and 2,874 testing data. No explicit mention of a separate validation split was found. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models, or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers. |
| Experiment Setup | Yes | For classification, we set Nℓto 256 and Ng to 32. When utilising k-NN for generating local-scale patches and searching neighbours for intra-learning, we assign the number of local-scale patches kℓto 64 and the number of neighbours for intra-scale learning kintra to 32. Settings for segmentation are same, except that Ng and kintra are changed to 64 and 16, respectively. |