GaussianEditor icon indicating copy to clipboard operation
GaussianEditor copied to clipboard

On getting the code to run on Sept. 2024

Open jhuangBU opened this issue 1 year ago • 10 comments

I just tried setting this entire project up again for something else, and the documented way of setting up conda environments appears to be broken (again).

Assuming the first way of setting up environments (by conda env create -f environment.yaml), you now need to do this instead:

  • comment out line 20 & line 21 in environment.yaml
    • we need to run pip manually
  • run conda env create -f environment.yaml
    • To setup the environment
  • run conda activate GaussianEditor
    • activate the environment for everything follows
  • run export CPLUS_INCLUDE_PATH=$CONDA_PREFIX/targets/x86_64-linux/include
    • so the NVCC compiler can find the cub/* thrust/* headers
  • go to gaussiansplatting/submodules/simple-knn/simple_knn.cu, and add #include <float.h> to the end of the include operations
    • so you won't get the FLT_MAX missing error during compilation
  • run pip install -r requirements.lock.txt --use-deprecated=legacy-resolver
    • install pip dependencies using the legacy solver so everything can still be found

All of this is very temporary and is only tested to work on 2024-09-09.

Yeah this is absolutely worth my time and is totally why I got into AI research! /s

jhuangBU avatar Sep 09 '24 18:09 jhuangBU

Thank you worked for most of the submodules but not for diff-gaussian-rasterizer,simpleknn,tinycudann. The error is with the compatibility of MSVS and Cuda toolkit. I have MSVS 2022 and cuda 11.8 and for these packages it says incompatible compiler. When i install gaussian splatting i had cuda 12.6 and MSVS 2022 and there was no error. Is there a workaround for this?

RAF3DA avatar Oct 04 '24 10:10 RAF3DA

Thank you worked for most of the submodules but not for diff-gaussian-rasterizer,simpleknn,tinycudann. The error is with the compatibility of MSVS and Cuda toolkit. I have MSVS 2022 and cuda 11.8 and for these packages it says incompatible compiler. When i install gaussian splatting i had cuda 12.6 and MSVS 2022 and there was no error. Is there a workaround for this?

I think there is a switch you can use to force nvcc to use the unsupported compiler, but otherwise you might have to fiddle with versions

jhuangBU avatar Oct 13 '24 17:10 jhuangBU

Made it work somehow, there is still a dependency conflict with Lang sem, it requires python>3.11. how to solve this?

RAF3DA avatar Oct 15 '24 12:10 RAF3DA

Then I suppose you need to fiddle with the versions a bit more; all of this was tested on Linux, sorry.

jhuangBU avatar Oct 15 '24 12:10 jhuangBU

Ok, for anyone having the same issue install lang-sem from this commit pip install git+https://github.com/luca-medeiros/lang-segment-anything.git@9bdf77715fa70fca96452184abdc64a6666e0d46

@jhuangBU Great work! Thank you!

RAF3DA avatar Oct 15 '24 13:10 RAF3DA

Ok, for anyone having the same issue install lang-sem from this commit pip install git+https://github.com/luca-medeiros/lang-segment-anything.git@9bdf77715fa70fca96452184abdc64a6666e0d46

@jhuangBU Great work! Thank you!

The link does not work now. It seems that the author of lang-sem deleted the branch and I failed to install the new version of lang-sem due to a version issue of accelerate.

jcwang-gh avatar Oct 17 '24 13:10 jcwang-gh

Try this one: https://github.com/luca-medeiros/lang-segment-anything/commit/9bdf77715fa70fca96452184abdc64a6666e0d46

RAF3DA avatar Oct 17 '24 13:10 RAF3DA

试试这个: luca-medeiros/lang-segment-anything@9bdf777

The latest lang-segment-anything requires python>3.11, but this causes dependency conflicts in igl. How can I solve it?The link above is also broken, how can we fix it? Thank you! Thank you!

husky2001 avatar Nov 11 '24 07:11 husky2001

Try this one: luca-medeiros/lang-segment-anything@9bdf777

i have the same problem, too. Is there any methods to solve it?

sgtziggy avatar Nov 25 '24 08:11 sgtziggy

Try this one: luca-medeiros/lang-segment-anything@9bdf777

i have the same problem, too. Is there any methods to solve it? This is version 0.1.0 I found, I succeeded, you can try it! https://github.com/paulguerrero/lang-sam

husky2001 avatar Nov 25 '24 08:11 husky2001