Installation Troubleshooting#

Solutions to the most common installation errors. If yours is not listed here, see Any other error.

Command not found#

If idtrackerai_test (or any other idtracker.ai command) fails right after installing with a short error like No such file or directory or Not recognized command, reactivate the Conda environment:

conda deactivate
conda activate idtrackerai

No graphic device was found available#

If your computer has an NVIDIA or AMD GPU, or runs macOS 12.3 or later on Apple silicon or an AMD GPU, but idtracker.ai does not find it, your PyTorch installation is not using the GPU.

Uninstall the whole Conda environment and install again, reading the PyTorch instructions for your computer carefully. Getting PyTorch to use the GPU can be tricky.

Issues with Qt#

idtracker.ai’s graphical apps use the PyQt6 Qt binding by default. If PyQt6 does not work on your computer, switch to another Qt binding:

  1. Uninstall PyQt6:

    python -m pip uninstall PyQt6

  2. Install one of the other Qt bindings:

    Alternative Qt bindings for Python#
    # choose one of the following options
python -m pip install PyQt5
python -m pip install PySide6
python -m pip install PySide2

  3. If idtracker.ai is not installed yet, install it without its dependencies, so that PyQt6 is not installed again, and then install the other dependencies by hand:

    # install idtracker.ai without any dependency
python -m pip install idtrackerai --no-deps
# install the rest of the dependencies, except PyQt6
python -m pip install numpy rich h5py scipy opencv-python-headless qtpy superqt toml matplotlib packaging psutil scikit-learn requests deprecated

idtracker.ai uses the first Qt binding it finds.

Note

If no Qt binding works on your computer, you can still track from a TOML file without the graphical interface, using the --track flag (see Terminal usage).

qt.qpa.plugin: Could not load the Qt platform plugin “xcb”#

On Ubuntu, install the missing library:

sudo apt install libxcb-cursor0

For other systems, see this thread.

Could not load library libcudnn#

If idtrackerai_test starts but fails after a few seconds with an error like Could not load library libcudnn_cnn_infer.so.8 (or any other libcudnn or libcuda library), PyTorch cannot find its GPU libraries.

PyTorch installed with pip includes its own CUDA and cuDNN libraries, so you do not need to install the CUDA Toolkit. The error usually means one of these:

  • PyTorch is outdated or broken. Reinstall it with the command from the PyTorch website (step 3 of Install idtracker.ai), adding --force-reinstall and --upgrade. An error mentioning .so.8 comes from an old PyTorch version that used cuDNN 8.

  • Other CUDA libraries take precedence (Linux). A CUDA installation on the system can shadow the libraries of PyTorch if it is listed in LD_LIBRARY_PATH. Check it with echo $LD_LIBRARY_PATH and, if it lists CUDA folders, test again in a terminal where it is cleared:

    unset LD_LIBRARY_PATH
idtrackerai_test

If none of this works, Uninstall the Conda environment and install idtracker.ai again from scratch.

module ‘torch’ has no attribute ‘_six’#

On macOS, upgrading PyTorch usually fixes this:

python -m pip install --upgrade torch

ImportError: cannot import name COMMON_SAFE_ASCII_CHARACTERS from charset_normalizer.constant#

Install chardet (from Stack Overflow):

python -m pip install chardet

Any other error#

Send us the error message, or the idtrackerai.log file, at info@idtracker.ai and we will help you.