Windows#
To install LabGym on Windows, you will need to access the terminal. To do this,
open the start menu by clicking the Win
key, type “PowerShell”, and hit
enter. All terminal commands going forward should be entered in this terminal.
Warning
There is a known issue with LabGym that prevents it from running on Windows 11. If you are on Windows 11, we suggest using Windows Subsystem for Linux (WSL) and following along with the Linux installation instructions.
Install Git.
If you’re unsure of which installation method to use, select the
64-bit Git for Windows Setup
option. Run the installer, and accept all default values.Install the Visual Studio C++ Build Tools.
Scroll down to the entry that says
Build Tools for Visual Studio 2022
and click “Download”.When you run the downloaded executable, you will be prompted to choose what tools you will need. Select only the
Desktop Development With C++
workload, then click “Install”.Install Python 3.10. Scroll down to the bottom and click the
Windows installer (64-bit)
option. Run the installer and accept all default options.To test your Python installation, run the following command. If the version number prints out successfully, your Python installation is working.
> py -3.10 --version Python 3.10.10
If you’re using an Nvidia GPU, install CUDA Toolkit 11.8 and cuDNN.
First, install and/or update your GPU drivers at this link. Select your GPU model and click “Search”, then click “Download”. After installing the drivers, reboot your system to ensure they take effect.
Then, install CUDA Toolkit 11.8. Select your version of Windows, select “exe (local),” then click “Download.”
Warning
If you’re using Windows Subsystem for Linux (WSL), please refer to the Linux install instructions.
To verify your installation of CUDA, use the following command.
> set CUDA_HOME=%CUDA_HOME_V11_8% > nvcc --version nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2022 NVIDIA Corporation Built on Wed_Sep_21_10:41:10_Pacific_Daylight_Time_2022 Cuda compilation tools, release 11.8, V11.8.89 Build cuda_11.8.r11.8/compiler.31833905_0
Finally, install cuDNN. You will need to register an Nvidia Developer account, which you can do for free.
Important
If you’re using Windows 11, when installing cuDNN, select “Tarball” then “11” under CUDA Version. Then, follow these instructions to install cuDNN from the
.tar.gz
file.As of February 2024, the latest version is cuDNN 9.0.0, which is compatible with CUDA 11.8.
Install
pipx
by following these instructions.To test your installation of
pipx
, close and reopen your terminal window, then type the following command:> pipx --version 1.4.3
If the version number prints successfully, then your installation is working properly. Otherwise, try running the
pipx ensurepath
command again.Install LabGym via
pipx
.> pipx install --python 3.10 LabGym
Install PyTorch in LabGym’s virtual environment.
> pipx inject --index-url https://download.pytorch.org/whl/cu118 LabGym torch==2.0.1 torchvision==0.15.2
If you are using LabGym without a GPU, use the following command instead.
> pipx inject --index-url https://download.pytorch.org/whl/cpu LabGym torch==2.0.1 torchvision==0.15.2
Install Detectron2 in LabGym’s virtual environment.
First, download the Detectron2 code using the following command.
> git clone https://github.com/facebookresearch/detectron2.git
The code should be available in the
C:\Users\<username>\detectron2\
folder.If you’re using a GPU, open the
setup.py
file inside thedetectron2
folder using a text editor (e.g. Notepad, VS Code, etc.). Go to line 79 in the code, then apply the following edit.Old version:
72if not is_rocm_pytorch: 73 define_macros += [("WITH_CUDA", None)] 74 extra_compile_args["nvcc"] = [ 75 "-O3", 76 "-DCUDA_HAS_FP16=1", 77 "-D__CUDA_NO_HALF_OPERATORS__", 78 "-D__CUDA_NO_HALF_CONVERSIONS__", 79 "-D__CUDA_NO_HALF2_OPERATORS__", 80 ] 81else: 82 define_macros += [("WITH_HIP", None)] 83 extra_compile_args["nvcc"] = []
New version:
72if not is_rocm_pytorch: 73 define_macros += [("WITH_CUDA", None)] 74 extra_compile_args["nvcc"] = [ 75 "-O3", 76 "-DCUDA_HAS_FP16=1", 77 "-D__CUDA_NO_HALF_OPERATORS__", 78 "-D__CUDA_NO_HALF_CONVERSIONS__", 79 "-D__CUDA_NO_HALF2_OPERATORS__", 80 "-DWITH_CUDA", 81 ] 82else: 83 define_macros += [("WITH_HIP", None)] 84 extra_compile_args["nvcc"] = []
Save the
setup.py
file, then exit your text editor.Finally, reopen your terminal, then install Detectron2.
> set CUDA_HOME=%CUDA_HOME_V11_8% > pipx runpip LabGym install -e detectron2
Launch LabGym.
> LabGym
The GUI will take a few minutes to start up during the first launch. If the LabGym GUI shows up, you have successfully installed LabGym!