yihua7 / SC-GS

[CVPR 2024] Code for SC-GS: Sparse-Controlled Gaussian Splatting for Editable Dynamic Scenes
https://yihua7.github.io/SC-GS-web/
MIT License
428 stars 21 forks source link

A few packages missed in requirements.txt, and share the experience in the installation #15

Closed liulisixin closed 3 months ago

liulisixin commented 3 months ago

First, thanks to the authors for creating this incredible project!

I've noticed that a few packages are missing from requirements.txt, which are necessary for the installation process. Adding these packages will ensure that the project runs smoothly:

To prevent train_gui.py from failing, please install: pip install lpips pip install pytorch_msssim

Additionally, to avoid runtime errors when interacting with the GUI, install: pip install matplotlib pip install scikit-image

Additionally, here is a collection of my installation commands:

2004 conda create --name scgs python=3.8 2005 conda activate scgs 2006 cd PycharmProjects/SC-GS/ 2007 pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116 2008 pip install -r requirements.txt (Remove torch from the original txt) 2010 export TORCH_CUDA_ARCH_LIST="8.6+PTX" (This is for the submodules because the setup cannot find GPU info on some machines. This number works for 3090 , check the number for other GPUs. ) (The related error info: arch_list[-1] += '+PTX' IndexError: list index out of range) 2011 pip install ./submodules/diff-gaussian-rasterization 2012 pip install ./submodules/simple-knn 2014 pip install lpips 2016 pip install pytorch_msssim 2019 pip uninstall pytorch3d 2022 FORCE_CUDA=1 pip install "git+https://github.com/facebookresearch/pytorch3d.git@stable" (FORCE_CUDA=1 for forcing the GPU version) (The related error info: pytorch3d, RuntimeError: Not compiled with GPU support.) pip install matplotlib pip install scikit-image

In addition, I work on Cuda 11.8 and Linux Ubuntu 20.04.

yilan120 commented 3 months ago

Thanks for your info, it helps a lot!

yihua7 commented 3 months ago

Hi, Many thanks for your kind reminder! I've updated the requirements.txt according to your feedback in the commit 1f835596c85422bb517df8c534cf8aaa6357ac5e. It will help a lot for all users. : )

liulisixin commented 3 months ago

Thanks for your info, it helps a lot!

Great!

liulisixin commented 3 months ago

Hi, Many thanks for your kind reminder! I've updated the requirements.txt according to your feedback in the commit 1f83559. It will help a lot for all users. : )

Thank you for your amazing project and work!

azzarelli commented 3 months ago

Hi, Thank you to the authors for the code release&paper and thank you to @liulisixin for the solution.

I ran this with cudatoolkit=11.8 and encountered "illegal access memory" error with the gaussian renderer.

This solution resolved my problems