question about training

[ch1998]

question about training

[tsabbir96]

Hi @ch1998 , were you successful in running the code? I mean I wanted to test the capabilities by giving prompt and checking what it generated. but the main issue is running the following command starts a training session which I dont want:
python launch.py --config configs/gaussiandreamer-sd.yaml --train --gpu 0 system.prompt_processor.prompt="a fox"

Did you find a way to test the model out-of-the-box without training it ?

[ch1998]

Sorry, I completed the training but didn't find a way to test it directly.

[tsabbir96]

@ch1998 Ah great how long did it take to complete the training?
Could you share the steps you used to install the code, I found that the steps mentioned in the Get Started section doesnot work if ran step by step

[ch1998]

Training takes about 10 minutes.
install step：
git clone https://github.com/hustvl/GaussianDreamer.git
cd GaussianDreamer
pip install torch==2.0.1+cu118 torchvision==0.15.2+cu118 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
pip install ninja
Modify xformers==0.0.20 in requirements.txt
pip install -r requirements.txt

pip install ./gaussiansplatting/submodules/diff-gaussian-rasterization
pip install ./gaussiansplatting/submodules/simple-knn
git clone https://github.com/openai/shap-e.git
cd shap-e
pip install -e .

pip install open3d
pip install plyfile

Please refer to #2 .

[tsabbir96]

@ch1998 great thanks... just one question i saw you have requirements.txt file even before cloning the GaussianDreamer Repo..
THe only requirements.txt file I saw is inside the repo.. Where did you get the requirements.txt before cloning the repo?

[ch1998]

@ch1998 great thanks... just one question i saw you have requirements.txt file even before cloning the GaussianDreamer Repo.. THe only requirements.txt file I saw is inside the repo.. Where did you get the requirements.txt before cloning the repo?

first，git clone https://github.com/hustvl/GaussianDreamer.git
then，other steps.
Sorry, the steps are in the wrong order.

[tsabbir96]

@ch1998 thank you so much for your help!!!

[tsabbir96]

@ch1998 could you tell me which GPU you used to reduce the training time to 10 mins, I am using T4 GPU on COlab and the time it takes is rediculously long.

[ch1998]

I am using a 3090. The first run will take some time. Some shape-e related files will be generated, which will take about an hour. Then there will be formal training, which will take about 10 minutes.

[CuzKy]

Hello, I successfully ran the code, but when the training reached normal level, the program did not continue. Is there any error? There is stabilityai/stable-diffusion-2-1-base in gaussiandreamer-sd.yaml. Is it because I did not download the pre-trained model?

i met the same problem.I encountered this situation when running this code on both RTX3090 and RTX3080. When I reached this state, there was no next step. At most, I ran it for almost 4 hours without taking any further steps, and my memory status did not change during the run.



how can i sovel it?thanks!!!!!!!