demo.ipynb: renderer disconnected on colab

[wayner0628]

I follow the instruction on demo.ipynb, but the viewer seems to be broken, both custom data and pre-built dozer fail. The former blocks executed correctly, and it did output all processed data. Due to the issue of viewer, I also could't train the model, it stopped at saving config. Not knowing why these happen, please check the issue, thanks!

[FuturaTino]

我和你遇到了一样的问题。这是由于colab目前默认pyhton版本3.10，tinycudann包版本是3.9tinycudann-1.7-cp39-cp39-linux_x86_64.whl，两者没有对应上，从而代码无法跑通。 解决方案如下： Colab界面 >Tools > Command palette > Use fallback runtime version 使用上一个python(也就是python3.9)，重启runtime，选择GPU设备，重新运行即可。

详情可见该文件：Quick fix for nerfstudio demo colab, issue #1077 https://github.com/vvdb101/nerfstudio_demo_issue_1077/blob/main/nerfstudio_demo_issue_1077.ipynb 希望能解决你的问题。

[loftusa]

Hi, unless i'm misunderstanding something, the renderer is disconnected for me as well.

[ichsan2895]

Seems similar issue with https://github.com/nerfstudio-project/nerfstudio/issues/1836 due to Google Colab upgrading their Python version from 3.9 to 3.10

[FuturaTino]

Seems similar issue with #1836 due to Google Colab upgrading their Python version from 3.9 to 3.10

Yeah. Using the fallback version of python will fix this.

[sudoku-lord]

I'm having the same issue in Kaggle, but downgrading my Python version there leads me into a fresh batch of errors (there's no super easy workaround in Kaggle to downgrade from 3.10 to 3.9 afaik). Is there a fix which can be implemented by nerfstudio itself? Or an updated version of tiny-cuda-nn which does work with 3.10?

[sudoku-lord]

As an update: I switched to Google Colab and found that the fallback runtime is no longer available, so I'm using this

!apt-get install python3.9
!ln -sf /usr/bin/python3.9 /usr/bin/python

before running the rest of my code, which has no errors. However, the renderer still shows up as disconnected. Can anyone render anything in the demo? I'd really love to get some insight on this, because I've been trying to get it working for quite a while now.

[Hawaiii]

Having the same problems (renderer disconnected, training doesn't start without any error message) and since the fallback runtime is no longer available, there seems to be no workaround for this issue at the moment.

[sudoku-lord]

@Hawaiii I have the same issue too, and posted about it here. I'd really really love some clarity on what's going on, because I haven't gotten any updates on it either.

[Eatonasher]

Having the same problems (renderer disconnected, training doesn't start without any error message) and since the fallback runtime is no longer available, there seems to be no workaround for this issue at the moment.

@Hawaiii is this still bugging up for you? I'm trying to make it work but everything keeps disconnecting.

[Hawaiii]

@Eatonasher @sudoku-lord I found out that if I installed the right version of tinycudann then this problem goes away. It’s annoying that each time I start the notebook tinycudann takes ~15min to install though.

I replace the tinycudann installation with the line below

!pip install ninja git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch