Semi-supervised 3D Object Detection with PatchTeacher and PillarMix
Authors: Xiaopei Wu, Liang Peng, Liang Xie, Yuenan Hou, Binbin Lin, Xiaoshui Huang, Haifeng Liu, Deng Cai, Wanli Ouyang
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on Waymo and ONCE datasets verify the effectiveness and superiority of our method and we achieve new state-of-the-art results, surpassing existing methods by a large margin. |
| Researcher Affiliation | Academia | 1State Key Lab of CAD&CG, Zhejiang University 2Shanghai AI Laboratory 3School of Software Technology, Zhejiang University {wuxiaopei, pengliang}@zju.edu.cn |
| Pseudocode | No | The paper describes the methods in prose and includes diagrams but does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Codes are available at https://github.com/LittlePey/PTPM. |
| Open Datasets | Yes | Waymo (Sun et al. 2020) is a large-scale Li DAR point cloud dataset, which contains 798 sequences for training and 202 sequences for validation." and "ONCE (Mao et al. 2021) is a large-scale autonomous driving dataset with 1 million Li DAR point cloud samples. |
| Dataset Splits | Yes | Waymo (Sun et al. 2020) is a large-scale Li DAR point cloud dataset, which contains 798 sequences for training and 202 sequences for validation." and "ONCE (Mao et al. 2021) is a large-scale autonomous driving dataset with 1 million Li DAR point cloud samples. There are 15k labeled samples, which are divided into 5K for training, 3k for validation and 8k for testing. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models used for running its experiments. |
| Software Dependencies | No | The paper mentions using "SECOND (Yan, Mao, and Li 2018) implemented by Open PCDet (Team 2020) as our baseline detector" but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | Yes | For the training of Patch Teacher, each mini-batch consists of 2 patches of labeled point clouds and 2 patches of unlabeled point clouds. We divide the full point clouds into 4 4 patches. The voxel size of each patch is set to [3.5 cm, 3.5 cm, 3.5 cm]. Patch Teacher is trained for 240 epochs. For the training of the student, each mini-batch consists of 1 frame of labeled point clouds and 4 frames of unlabeled point clouds. The student is trained for 30 epochs. |