Geometry-Aware 3D Salient Object Detection Network
Authors: Chen Wang, Liyuan Zhang, Le Hui, Qi Liu, Yuchao Dai
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that our method achieves new state-of-the-art performance on the PCSOD dataset. ... To verify the effectiveness of the proposed method, we conduct experiments on the point cloud salient object detection dataset (PCSOD) (Fan, Gao, and Li 2022). Extensive experiments have shown that our approach is significantly superior to other methods and has a shorter running time. ... We conduct a comparative analysis of performance by comparing our method with three state-of-the-art point cloud salient object detection methods... To analyze the effectiveness of each proposed module, we conduct comprehensive ablation study experiments on the PCSOD dataset. |
| Researcher Affiliation | Academia | 1Shaanxi Key Laboratory of Information Acquisition and Processing, Northwestern Polytechnical University 2PCA Lab, Key Lab of Intelligent Perception and Systems for High-Dimensional Information of Ministry of Education, Nanjing University of Science and Technology EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: Superpoint Generation Algorithm Input: Unclustered point set U, Queue size K, Distance threshold γ Output: Superpoint sets S 1 while U = do 2 Randomly select a point i U as cluster center; 3 Initialize queue Q as empty; 4 for each point j U \ {i} do 5 Compute Euclidean distance dij ; 6 Enqueue (j, dij) into Q; 7 Sort Q by distance and keep the K nearest points; 8 while Q is not empty do 9 Extract point j from Q; 10 Compute feature distance d(i, j); 11 if d γ then 12 Add j to the superpoint of i; 13 Remove j from U; 15 Clear queue Q; 17 return S; |
| Open Source Code | No | The paper does not provide an explicit statement about releasing code or a link to a code repository. |
| Open Datasets | Yes | Following previous methods (Fan, Gao, and Li 2022; Wei et al. 2024; Zhang et al. 2023), we adopt the point cloud salient object detection dataset (PCSOD) (Fan, Gao, and Li 2022) as the benchmark to conduct experiments. |
| Dataset Splits | Yes | There are a total of 2,873 scenarios in this dataset, including simple targets, multiple targets, small targets, and some targets with complex structures. According to the official partition, there are 2,000 samples for training and 872 samples for evaluation. |
| Hardware Specification | Yes | Our method is implemented with the Py Torch on an NVIDIA RTX 3090 Ti GPU. |
| Software Dependencies | No | Our method is implemented with the Py Torch on an NVIDIA RTX 3090 Ti GPU. (Only 'Py Torch' is mentioned, not with a specific version number, and no other software dependencies with versions are listed). |
| Experiment Setup | Yes | For the training process, our model is trained using the Adam optimizer for a total of 300 epochs, taking approximately 13 hours. ... The initial learning rate is 1e-3 and weight decay is 1e-4. Additionally, data augmentation is applied using transformation with standard deviations of [3, 3, 3]. The parameter spatial shape is set to [150, 100, 75], indicating the shape of the spatial grid with dimensions 150, 100, and 75 along the x, y, and z axes, respectively. |