Title: Multiple Installation Issues Encountered with SoccerNeRF
Hello Sacha
I've been trying to run the SoccerNeRF code from your paper but have encountered multiple issues that prevent successful installation and execution. I followed the setup instructions provided in the README at the SoccerNeRF repository.
Environment:
- OS: Linux
- Python Version: 3.8.19
- tinycudann Version: 1.7
- CUDA Version: 11.7
- PyTorch: 2.0.1 (I tried it first with 1.13.1 it was throwing an error due to the CUDA version)
Steps to Reproduce:
- Cloned the repository
- Ran
ns-install-clias per the instructions in the directory mentioned above.
Expected Behavior:
The installation completes without errors, and the ns-train command executes as described.
Actual Behavior:
Encountered errors during installation and when running the ns-train command. Below are the key issues:
- Running
ns-install-cliresulted in multiple errors, which seem related to dependencies or script execution issues. - Trying to execute the
ns-traincommand with the following parameters:
ns-train k-planes \
--vis viewer \
--pipeline.model.multiscale-res 1 2 4 8 16 32 \
--pipeline.datamanager.ist-range 0.75 \
broadcaststyle-data \
--fps-downsample 4
This resulted in errors indicating that several arguments do not exist.
Screenshots:
Could you please provide some guidance on how to resolve these issues? Any help would be greatly appreciated.
Many Thanks.
Hello Mustafa,
Sorry for not seeing your issue and email earlier!
It seems you've installed Nerfstudio separately? You mention these steps:
- Cloned the repository
- Ran ns-install-cli as per the instructions in the directory mentioned above.
Did you follow the intermediary steps?
cd nerfstudio
pip install --upgrade pip setuptools
pip install -e .
pip install -e .[dev]
I recommend making a clean Conda environment, and starting from scratch and following all the instructions in the README. Your version of Nerfstudio seems not to be the one of the paper but rather the recent one from Nerfstudio itself, which has become quite different.
Indeed, in your version, there is the other version of K-Planes, called kplanes with no dash. You can try using kplanes with no dash, but you won't be able to use importance sampling.
Sadly, I was not able to merge all my changes into Nerfstudio, and I think someone implemented some sort of importance sampling (IST and/or ISG) but I'm not sure. I had to switch projects and could only implement part of the changes, which is why I let my code at the time in this repository for reproducibility.
To run the data on new versions of Nerfstudio, you will need to manually adapt the datamanagers/parsers of nerf studio by following the ones in this repository.
I hope this helps you, Best Regards, Sacha Lewin