Pamba: Enhancing Global Interaction in Point Clouds via State Space Model
Authors: Zhuoyuan Li, Yubo Ai, Jiahao Lu, ChuXin Wang, Jiacheng Deng, Hanzhi Chang, Yanzhe Liang, Wenfei Yang, Shifeng Zhang, Tianzhu Zhang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Pamba obtains state-of-the-art results on several 3D point cloud segmentation tasks, including Scan Net v2, Scan Net200, S3DIS and nu Scenes, while its effectiveness is validated by extensive experiments. We conduct extensive experiments and ablation studies to validate our design choices. Pamba achieves state-of-the-art performance on several highly competitive point cloud segmentation tasks, including Scan Net v2, Scan Net200, S3DIS, and nu Scenes. |
| Researcher Affiliation | Collaboration | 1Deep Space Exploration Laboratory/School of Information Science and Technology, University of Science and Technology of China 2Sangfor Technologies Inc. |
| Pseudocode | No | The paper describes the proposed methods, including the multi-path serialization strategy and Conv Mamba block, using detailed textual explanations and architectural diagrams (Figure 2 and Figure 5). However, it does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not explicitly state that the source code for the methodology described is publicly available, nor does it provide a link to a code repository or mention code in supplementary materials. |
| Open Datasets | Yes | We evaluate Pamba on four datasets: Scan Net v2(Dai et al. 2017), Scan Net200 (Rozenberszki and Dai 2022), S3DIS (Armeni et al. 2016) and nu Scenes (Caesar et al. 2019; Fong et al. 2021). |
| Dataset Splits | Yes | Scan Net, Scan Net200 and nu Scenes follow the official data splits to generate corresponding training set, validation set and test set. For S3DIS, the data is split into six areas, the fifth of which is withheld during training and used for evaluation. |
| Hardware Specification | Yes | We train our model on 4 RTX 3090 GPUs. All measurements are taken on a single RTX 3090 |
| Software Dependencies | No | Adam W (Loshchilov and Hutter 2017) is adopted for parameter optimization. The paper mentions this optimizer but does not specify any other software dependencies with version numbers, such as programming languages or libraries. |
| Experiment Setup | Yes | We train our model on 4 RTX 3090 GPUs. Adam W (Loshchilov and Hutter 2017) is adopted for parameter optimization. Scan Net and Scan Net200 are trained 800 epochs, S3DIS is trained 3000 epochs and nu Scenes is trained 50 epochs. The sum of Cross Entropy loss and Lovasz loss (Berman, Triki, and Blaschko 2018) is adopted as the overall loss for Pamba as shown in Eq. 4, where LCE represents Cross Entropy loss and LL represents Lovasz loss. λ1 and λ2 are both empirically set to 0.5 during implementation. |