华东师范大学学报(自然科学版) ›› 2023, Vol. 2023 ›› Issue (6): 61-72.doi: 10.3969/j.issn.1000-5641.2023.06.006

• 计算机科学 • 上一篇    下一篇

动量更新与重构约束的限制视角下3D物品识别

崔瑞博, 王峰*()   

  1. 华东师范大学 计算机科学与技术学院, 上海 200062
  • 收稿日期:2022-06-18 出版日期:2023-11-25 发布日期:2023-11-23
  • 通讯作者: 王峰 E-mail:fwang@cs.ecnu.edu.cn

Momentum-updated representation with reconstruction constraint for limited-view 3D object recognition

Ruibo CUI, Feng WANG*()   

  1. School of Computer Science and Technology, East China Normal University, Shanghai 200062, China
  • Received:2022-06-18 Online:2023-11-25 Published:2023-11-23
  • Contact: Feng WANG E-mail:fwang@cs.ecnu.edu.cn

摘要:

提出了一个基于动量更新表示与重构约束的神经网络训练框架: 在视角标签信息缺失的限制性条件下, 使用2D (two-dimensional)图像进行3D (three-dimensional)物品识别. 首先, 使用自监督学习来解决训练过程中标签缺失的问题. 其次, 在动态队列基础上, 使用动量更新来保持物品表示的稳定性. 更进一步地, 在训练框架中加入自编码器模块, 利用重构约束使模型学习到的表示具有更多的语义信息. 最后, 提出动态队列递减策略, 解决训练过程中数据分布不均衡带来的准确度下降问题. 在2个广泛使用的多视角数据集ModelNet和ShapeNet上进行了实验, 结果表明所提方法具有良好的性能表现.

关键词: 多视角物品识别, 自监督学习, 自编码器

Abstract:

We propose a neural network training framework called momentum-updated representation with reconstruction constraint for 3D (three-dimensional) object recognition using 2D (two-dimensional) images without angle labels. First, self-supervised learning is employed to address the lack of angle labels. Second, we use momentum updating based on a dynamic queue to maintain the stability of the object representation. Furthermore, the reconstruction constraint is applied to the learning process with an auto-encoder module, which enables the representation to capture more semantic information of the objects. Finally, during training, a dynamic queue reduction strategy is proposed for handling the imbalanced data distribution. Experiments on two popular multi-view datasets, ModelNet and ShapeNet, demonstrate that the proposed method outperforms existing methods.

Key words: multi-view object recognition, self-supervised learning, auto-encoder

中图分类号: