基于正则化损失的 MeshNet 半监督分类

深度学习在网格分类中的应用越来越受到人们的关注, 在网格分类任务中, 通常使用交叉熵损失作为损失函数。文章提出一种利用数据的结构相似性和几何一致性的正则化损失, 将其加入损失函数中进行优化, 可有效提高网格的分类准确率。从实验结果的量化指标来看, 提出的正则化损失对于网格半监督分类任务的准确率有很好的提升效果。

The application of deep learning in mesh classification has attracted increasing attention. In mesh classification tasks, cross entropy loss is usually used as a loss function. In this paper, a regularization loss based on the structure similarity and geometric consistency of data is proposed, which can be directly added to the loss function to improve the classification accuracy of mesh. According to the final quantitative index of experimental results, the proposed regularization loss has a good effect on improving the accuracy of semi-supervised mesh classification task.