ingra14m/Deformable-3D-Gaussians

D:\anaconda3\envs\deformable_gaussian_env\lib\site-packages\torch\include\pybind11\cast.h(626): error: too few arguments for template template parameter "Tuple" detected during instantiation of class "pybind11::detail::tuple_caster<Tuple, Ts...> [with Tuple=std::pair, Ts=<T1, T2>]"

Lingbeaty opened this issue · 5 comments

Hi, I have been encountering an error when configuring the environment, and after searching the Internet, I found that VS, pytorch and CUDA are not suitable for this version, and I have tried various versions of VS, pytorch and CUDA and still have not solved it. I wonder if you can elaborate on the VS, pytorch and CUDA versions you use, thank you very much! Here's requirements.txt I did pip install -r
Error reported when reporting:
D:\anaconda3\envs\deformable_gaussian_env\lib\site-packages\torch\include\pybind11\cast.h(626): error: too few arguments for template template parameter "Tuple"
detected during instantiation of class "pybind11::detail::tuple_caster<Tuple, Ts...> [with Tuple=std::pair, Ts=<T1, T2>]"
(723): here

D:\anaconda3\envs\deformable_gaussian_env\lib\site-packages\torch\include\pybind11\cast.h(719): error: too few arguments for template template parameter "Tuple"
detected during instantiation of class "pybind11::detail::tuple_caster<Tuple, Ts...> [with Tuple=std::pair, Ts=<T1, T2>]"
(723): here

Direct machine translation have a bit problem, there is the code I want to run and encounter error:
pip install -r requirements.txt

Hi, thanks for your interest.

My CUDA version is 11.6, PyTorch is 1.13.1, and the operating system is Ubuntu. I suspect that your problem is likely caused by the installation of depth-diff-gaussian-rasterization.

I believe whether it's Ubuntu or Windows is extremely important. I have tested on a Debian system before and indeed failed to successfully install the environment.

Thank you very much! I rent a server today and it's Ubuntu system. I configured the environment successfully!

I think SC-GS is a better choice if you have limited computing resources. It has higher quality, less 3D Gaussians than Deformable-GS, and even faster than vanilla 3D-GS. Besides, I highly recommend Spacetime-GS and 4D-Gaussian-splatting (pku) for multi-view scenes.

Really thank you for your advice, I will have a try : )