An Tao
Hi everyone! I successfully modified the code to the spconv 2.3.6! See this [link](https://github.com/xinge008/Cylinder3D/issues/107#issuecomment-1592291831) for details.
Hi everyone! I successfully modified the code to the `spconv 2.3.6`! I can obtain the same computation results in every network layers as the provided code in this repo for...
Sorry, I don't understand how your classification works. In this repo, the model can do the classification task. The input is a point cloud of $N$ points in XYZ coordinate...
And what your batch size do you use? After I add `--dataset ScannetSparseVoxelization2cmDataset`, I find batch size 2 out of memory in my 2080ti.
I follow the original TensorFlow code in [here](https://github.com/WangYueFt/dgcnn/tree/master/pytorch). The architecture in this repo is the same as the original code.
The `sphere.npy` is created with an algorithm using spherical coordinates. 1. Randomly set a $\theta$ in $0 \sim 2\pi$ and a $\phi$ in $-\pi/2 \sim \pi/2$. 2. Obtain $x$, $y$,...
你可以增加点数，对于点数为$N$的一个点云，所需要内存消耗是$N^2$的倍数，所以你需要考虑你的机器计算内存是否够用。
下采样当然是可以的，但是你需要考虑到下采样后点云可能质量不佳，比如下采样后的点过于稀疏，分布不均，等等问题