Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding
Authors: Yanmei Zou, Hongshan Yu, Zhengeng Yang, Zechuan Li, Naveed Akhtar
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The developed technique is extensively evaluated for 3D object classification, object part segmentation, semantic segmentation and object detection. We establish new state-of-the-art results of 87.6 m Acc on Scan Object NN for object classification, and 85.5 class m Io U on Shape Net Part for object part segmentation, and 72.7 and 78.7 m Io U on Area-5 and 6-fold experiments with S3DIS for semantic segmentation. |
| Researcher Affiliation | Academia | Yanmei Zou1, Hongshan Yu1*, Zhengeng Yang2*, Zechuan Li1, Naveed Akhtar3 1College of Electrical and Information Engineering, Quanzhou Innovation Institute, Hunan University, Changsha, China 2College of Engineering and Design, Hunan Normal University, Changsha, China 3School of Computing and Information Systems, The University of Melbourne, 3052 Victoria, Australia |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code for this work is available at https://github.com/zouyanmei/HPENet. |
| Open Datasets | Yes | Scan Object NN (Uy et al. 2019) collects real-world objects from 700 unique scenes of the SOTA mesh datasets Scene NN (Hua et al. 2016) and Scan Net (Dai et al. 2017). S3DIS (Armeni et al. 2016) comprises 6 large-scale indoor areas and 271 rooms, which are captured from 3 different buildings. Scan Net V2 (Dai et al. 2017) consists of 3D indoor scenes with 2.5 million RGB-D frames in more than 1,500 scans, annotated with 20 semantic classes. |
| Dataset Splits | Yes | For evaluation, we use the popular metrics of mean Io U (m Io U), m Acc, and OA. From Tab. 2, it can be observed that HPENet establishes new state-of-the-art performances of 72.7% m Io U on S3DIS Area-5 and 78.7% m Io U on S3DIS (6-fold cross-validation). We follow the standard training and validation splits of 1,201 and 312 scenes, respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | We develop HPENets with the following configurations in our experiments. Scan Object NN: Ce = 32, B = [0, 0, 0 ,0]. Model Net40: Ce = 64, B = [0, 0, 0 ,0]. Shape Net Part: Ce = 160, B = [0, 0, 0 ,0]. S3DIS: Ce = 64, B = [3, 6, 3 ,3]. Scan Net V2: Ce = 64, B =[5, 8, 5 ,5]. Following Qi et al. (2017b), we randomly select 2,048 points as input and use class mean Io U (Cls. m Io U) and instance mean Io U (Ins. m Io U) for evaluation. |