Dockerfile？

[yuancaimaiyi]

@half-potato

Hello, very good work, but can you provide Dockerfile? I have been configuring cuda11.8 ubuntu20.04 on my computer for a whole day, and there are problems everywhere. So can you provide Docker image or Dockerfile file, so that everyone can quickly verify the code and spread the code more widely? Thank you

[sunwonlikeyou]

I wish it could be!

[half-potato]

I will try to put something together. Give me a bit of time. You will still need to build the SIBR viewer to view it.

[sunwonlikeyou]

Thank you for your efforts!. We have a many viewers like Supersplat.
So It will be thankful only for giving training enviorments.

[half-potato]

I have to find a workaround this issue:
shader-slang/slang#6432

Might take a bit of time

[yuancaimaiyi]

I have to find a workaround this issue:我必须找到解决这个问题的方法： shader-slang/slang#6432

Might take a bit of time可能需要一些时间

I am looking forward to it. I have the same problem with slang, and I can't install pytorch3d. I am looking forward to your dockerfile coming out, so that I won't waste time on environment configuration, and I will have more time to learn code.

[yuancaimaiyi]

@half-potato

Sorry to bother you, is there any progress on dockerfile? I have compiled all versions of slang again these days. Either slang compiles fine, but ever compiles with problems, or slang compiles with problems

[half-potato]

Sorry for the delay. I was able to get stuff compiling by downloading the slang release. However, now I'm stuck trying to get Optix into the Dockerfile. I can't really download it anywhere, and I'm currently working on a deadline.

[half-potato]

git clone --recursive https://github.com/half-potato/ever_training
bash docker_train.sh <path to dataset> <path to output>

And to visualize, do:

bash docker_visualize.py <path to output> <path to dataset> 6006 127.0.0.1

and in a different terminal:

cd SIBR_viewers
./install/bin/SIBR_remoteGaussian_app --ip 127.0.0.1 --port 6009

[sunwonlikeyou]

Thank you for your effort..!!
But I pulled your docker image splinetracer wasn't made.

Traceback (most recent call last):
File "/workspace/ever_training/train.py", line 17, in
from gaussian_renderer.ever import splinerender
File "/workspace/ever_training/gaussian_renderer/ever.py", line 20, in
from ever.splinetracers.fast_ellipsoid_splinetracer import trace_rays
File "/codes/2025/ever_training/ever/splinetracers/fast_ellipsoid_splinetracer.py", line 27, in
from build.splinetracer.extension import fast_ellipsoid_splinetracer_cpp_extension as sp
ModuleNotFoundError: No module named 'build.splinetracer'

[yuancaimaiyi]

@half-potato Thank you very much, but this problem occurs when I run it with the docker image you provided. I have no problem on another computer. It is a very strange error.
python train.py -h Traceback (most recent call last): File "/ever_training/train.py", line 17, in <module> from gaussian_renderer.ever import splinerender File "/ever_training/gaussian_renderer/ever.py", line 20, in <module> from ever.splinetracers.fast_ellipsoid_splinetracer import trace_rays File "/ever_training/ever/splinetracers/fast_ellipsoid_splinetracer.py", line 33, in <module> otx = sp.OptixContext(torch.device("cuda:0")) RuntimeError: OPTIX_ERROR_LIBRARY_NOT_FOUND: Optix call 'optixInit()' failed: /ever_training/ever/splinetracers/fast_ellipsoid_splinetracer/py_binding.cpp:80)

[half-potato]

I pushed another commit that should fix the error that @sunwonlikeyou faced. I'm not sure if it fixes @yuancaimaiyi's error, but can you check? If not, please post information about your machine. You need optix library installed on the host machine, I believe. And a CUDA enabled GPU, of course.

[sunwonlikeyou]

I pulled your latest image, and i ran the 'train.py' .

But I still got this messege

ModuleNotFoundError: No module named 'build.splinetracer'

[half-potato]

Not the latest image. The latest repo change. The difference is in the script, not the image.

[sunwonlikeyou]

Training starts!!
Thank you so much your effort.