您好，欢迎访问上海市农业科学院 机构知识库！
上海市农业科学院机构知识库

全部

  • 全部
  • 标题
  • 学者
  • 机构
  • 关键词

Impact of Pretraining Datasets on Point-MAE Performance in Point Cloud Classification

文献类型： 会议论文

第一作者： Rita Younes

作者： Rita Younes 1 ; Roy Abi Zeid Daou 1 ; Grace El Khoury 1 ; Jalal Possik 2 ; Charles Yaacoub 2 ;

作者机构： 1.Universite La Sagesse, Faculty of Engineering - Polytech, Biomedical Engineering Department

2.ICL, Junia, Universite Catholique de Lille, LITL

关键词： 3D modeling;Performance benchmarking;Point cloud classification;Point-MAE;Self-supervised learning

会议名称：

主办单位：

页码： 289-293

摘要： Point clouds are advantageous for their ability to represent three-dimensional structures and are widely used in applications such as 3D modeling, autonomous driving, and medical imaging. In machine learning, point cloud processing can be approached through self-supervised learning, unsupervised learning, and supervised learning methods. Currently, the number of self-supervised learning methods is limited, and their efficiency is often studied on single datasets, leaving the overall performance of these methods not well defined. This study evaluates the Point-MAE method for point cloud classification through practical experiments, extending beyond theoretical reviews. By pretraining the model on various datasets, including ShapeNet and ModelNet40, and evaluating its performance on ScanObjectNN's variants, we aimed to verify the performance of this method when dealing with different datasets. Our findings reveal slight discrepancies in accuracies, with ShapeNet pretraining achieving the highest accuracy for classification on ModelNet40 with 92.95% but requiring significantly more time compared to ScanObjectNN pretraining that reached an accuracy of 92.79%. This study highlights the critical impact of pretraining dataset selection on model performance and computational efficiency. The practical insights gained underscore the necessity of hands-on testing to fully understand the robustness and adaptability of self-supervised learning methods.

分类号： tp391.9-53

  • 相关文献
作者其他论文 更多>>
置顶
购物车
反馈

© 京ICP备09089781号-29 主办单位：上海市农业科学院

学术资源： 中国农业科学院 农业专业知识服务系统 农科联盟资源共建共享平台 中国农业科技文献与信息服务平台 中国农业期刊集成服务平台