Generalized Zero-Shot Learning for Point Cloud Segmentation with Evidence-Based Dynamic Calibration
Authors: Hyeonseok Kim, Byeongkeun Kang, Yeejin Lee
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that the proposed approach achieves state-of-the-art performance on generalized zero-shot semantic segmentation datasets, including Scan Net v2 and S3DIS. |
| Researcher Affiliation | Academia | Hyeonseok Kim, Byeongkeun Kang, Yeejin Lee* Seoul National University of Science and Technology, Republic of Korea EMAIL |
| Pseudocode | No | The paper describes its methodology in prose and mathematical formulations but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We evaluate the proposed E3DPC-GZSL method using the S3DIS dataset (Armeni et al. 2017) and the Scan Net v2 dataset (Dai et al. 2017), following the data splitting protocol outlined in previous studies (Michele et al. 2021; Yang et al. 2023a). |
| Dataset Splits | Yes | The Scan Net v2 dataset, collected indoors, consists of 1, 201 point cloud scenes for training and 312 point cloud scenes for evaluation. ... The S3DIS dataset consists of 272 scenes from 6 indoor areas, covering 13 classes. Areas 2, 3, 4, 5, and 6 (228 scenes) are used for training with nine seen classes (Ns = 9), while Area 1 with 44 scenes is designated as the evaluation dataset. |
| Hardware Specification | No | The paper does not specify any particular GPU models, CPU models, or other hardware components used for running the experiments. |
| Software Dependencies | No | We implement the proposed method using the Py Torch framework. ... We employ 600-dimensional (Nt = 600) text embeddings, which are a concatenation of Glo Ve (Pennington, Socher, and Manning 2014) and Word2Vec (Mikolov et al. 2013)... |
| Experiment Setup | Yes | The proposed model is trained with a learning rate of 7e 2, a batch size of 4, and a poly learning rate scheduler with a base of 0.9 and 30 epochs. The Adam optimizer is used for Scan Net v2, while the SGD optimizer is used for S3DIS. Each minibatch contains 8192 sample points (Nb = 8192). λDL and λBL are set to 0.005 and 0.01 for Scan Net v2 and 0.005 and 0.1 for S3DIS, respectively. |