First time streaming in the U.S. (Talking about Gaussian Splatting cuda code) [English CC] 

我最近也刚开始学习NeRF，零基础，要怎么入门呢

不知道是否可以再讲讲如果有球谐函数的反向传播怎么计算，直播里只讲了render这一步的反向传播，preprocess的感觉更复杂 不过讲的部分非常清晰非常感谢

還是你Gaussian的數量丟的不夠多？

您指的Gaussian是什么的数量？点云的数量吗？我是用的metashape获取的位姿，然后丢给高斯进行计算的@@sophist52

@@尚衫夏香 初始化的点云数量不够，我做室内场景发现有个问题是如果你初始化点太少，或者不在关键位置上，效果不好。我感觉你把图片丢到colmap做dense reconstruction，然后把dense points丢进去初始化，估计效果能不错

好的！我尝试一下！感谢分享！@@bookerml8502

added

先转mesh，mesh再转voxel会更容易。

你好，我也有这个问题。请问有解决的办法了吗

python 调用 cuda/c++ 实现，光栅化实现是单独的git仓库。在主仓库 submodules 目录下面。

on

the original code already does that, but it stores in some buffer that is very hard to extract from pytorch side. If you want to retrieve that info, you basically need to create a pytorch tensor, and pass the pointer to that tensor to the cuda code, then you can retrieve from python. It is rather complicated.

@@kwea123can you tell me which part of the code does it. I dont know CUDA much

​@@JaswinKasi github.com/graphdeco-inria/gaussian-splatting/issues/403 I pointed out in this issue, but it is rather complicated without code explanation.

@@kwea123 thanks!

把rgb的部分改成point的Z就可以了，但這個要另外自己用cuda寫，已經有人寫好了，可參考github.com/JonathonLuiten/diff-gaussian-rasterization-w-depth

@@kwea123 谢谢～