我们提出了一种三维点云分析的非参数网络Point-NN，它由完全不可学习的组件组成：最远点采样（FPS）、k近邻（k-NN）和池化操作，以及三角函数。令人惊讶的是，它在各种3D任务中表现良好，不需要参数或训练，甚至超越了现有的完全训练模型。从这个基本的非参数模型出发，我们提出了两个扩展。首先，Point-NN可以作为一个基础架构框架来构建参数化网络，只需在其上方插入线性层即可。由于优越的非参数基础，所得到的Point-PN表现出高性能效率折衷，只需要很少的可学习参数。其次，Point-NN可以被看作是已经训练好的3D模型的即插即用模块，在推理过程中捕获补充几何知识，增强现有方法对不同3D基准的表现，无需重新训练。我们希望我们的工作能够为社区提供关于用非参数方法理解三维点云的启示。代码可在https://github.com/ZrrSkywalker/Point-NN中获得。
We present a Non-parametric Network for 3D point cloud analysis, Point-NN,
which consists of purely non-learnable components: farthest point sampling
(FPS), k-nearest neighbors (k-NN), and pooling operations, with trigonometric
functions. Surprisingly, it performs well on various 3D tasks, requiring no
parameters or training, and even surpasses existing fully trained models.
Starting from this basic non-parametric model, we propose two extensions.
First, Point-NN can serve as a base architectural framework to construct
Parametric Networks by simply inserting linear layers on top. Given the
superior non-parametric foundation, the derived Point-PN exhibits a high
performance-efficiency trade-off with only a few learnable parameters. Second,
Point-NN can be regarded as a plug-and-play module for the already trained 3D
models during inference. Point-NN captures the complementary geometric
knowledge and enhances existing methods for different 3D benchmarks without
re-training. We hope our work may cast a light on the community for
understanding 3D point clouds with non-parametric methods. Code is available at
https://github.com/ZrrSkywalker/Point-NN.
论文链接：http://arxiv.org/pdf/2303.08134v1
原创文章，作者：fendouai，如若转载，请注明出处：https://panchuang.net/2023/03/15/%e5%8f%82%e6%95%b0%e5%b9%b6%e4%b8%8d%e6%98%af%e6%82%a8%e6%89%80%e9%9c%80%e8%a6%81%e7%9a%84%e5%85%a8%e9%83%a8%ef%bc%9a%e4%bb%8e%e9%9d%9e%e5%8f%82%e6%95%b0%e7%bd%91%e7%bb%9c%e5%bc%80%e5%a7%8b%e8%bf%9b/
fendouai 