TANG Zhiye

Ok after I turned down the no match for `zsh`, and then switched to tf-1.15, I could finally build horovod for TensorFlow. But when I tried again to `import horovod.tensorflow`,...

Yes, I've installed tf-1.x via Anaconda and it seems that nothing goes wrong.

Just via a command in `zsh`: `pip install tensorflow-1.x`, and the installation goes well. I don't exactly know how to debug by adding a `-v`? Can you offer some related...

Is simply setting `gaussians.max_radii2D` as all-zero tensors like in `create_from_pcd()`, as ```{Python} self.max_radii2D = torch.zeros((self.get_xyz.shape[0]), device="cuda") ``` feasible? I'm not sure.

> From my knowledge, I can say that  > > `self.max_radii2D` > > are used when projecting 3D gaussians to a 2D surface. Yes,  `max_radii2D` is set to be all...

@luoww1992 Maybe you can try pruning all view-dependent information used in Octree-GS and re-train the code. Afterward, you can use the way as in `generate_neural_gaussians()` to generate all neural Gaussians...

The fundamental reason why Scaffold-GS derivatives (like Octree-GS and HAC) are difficult to import into UE is because they require generating neural Gaussians on-the-fly every time the view changes. Therefore,...

What‘s more, I've noticed that some anchors **do not** have a parent anchor.

I modified your code to prevent duplicate anchors from growing. Afterward, I re-trained `bicycle` on your released code, finding that there are numerous ``orphan'' anchors with no parent anchors, and...

The existence of orphan anchors might be due to the growing strategy, which does not consider whether the candidates' parents exist. As to the latter, I don't understand why the...