Unified 3D Segmenter As Prototypical Classifiers

Authors: Zheyun Qin, Cheng Han, Qifan Wang, Xiushan Nie, Yilong Yin, Lu Xiankai

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical results demonstrate that PROTOSEG outperforms concurrent well-known specialized architectures on 3D point cloud benchmarks, achieving 72.3%, 76.4% and 74.2% m Io U for semantic segmentation on S3DIS, Scan Net V2 and Semantic KITTI, 66.8% m Cov and 51.2% m AP for instance segmentation on S3DIS and Scan Net V2, 62.4% PQ for panoptic segmentation on Semantic KITTI, validating the strength of our concept and the effectiveness of our algorithm.
Researcher Affiliation Collaboration Zheyun Qin1 , Cheng Han2 , Qifan Wang3, Nie Xiushan4, Yilong Yin1 , Xiankai Lu1 1Shandong University, 2Rochester Institute of Technology, 3Meta AI, 4Shandong Jianzhu University
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes The code and models are available at https://github.com/zyqin19/PROTOSEG.
Open Datasets Yes S3DIS [9], a large-scale indoor point cloud dataset, encompasses point clouds from 271 rooms across 6 areas." "Scan Net V2 [10] provides over 1,500 indoor scenes and around 2.5 million annotated RGB-D images..." "Semantic KITTI [11] is introduced based on the well-known KITTI Vision [71] benchmark...
Dataset Splits Yes Table 2: Comparisons of semantic segmentation with m Io U on Scan Net v2 [10] (see 5.1). Method Test Val. and Table 3: Comparisons of semantic segmentation performance on Semantic KITTI val set (see 5.1).
Hardware Specification Yes Training and testing are conducted on eight NVIDIA A100 GPUs.
Software Dependencies No The paper does not explicitly provide specific software dependencies with version numbers, such as Python library versions or framework versions.
Experiment Setup Yes The hyper-parameter κ balances the convergence speed and stability of Eq. 10 in addition to smoothing the association (Eq. 9). We just use κ = 0.05 following [30] for our experiments, not extensively fine-tuned." and "Our model achieves the best performance when the momentum coefficient is set to 0.999." and "The m Io U score peaks at 72.34% when K = 10."