Visitar URL original
DeepLabCut YAML install works on macOS, but on Windows the GUI fails with Qt/PySide6 import errors · Issue #3252 · DeepLabCut/DeepLabCut · GitHub
Skip to content

DeepLabCut YAML install works on macOS, but on Windows the GUI fails with Qt/PySide6 import errors #3252

Closed

Description

@RaphJP

Hi, I’m trying to set up DeepLabCut on a Windows workstation for behavior analysis, and I ran into a GUI issue when installing from the YAML environment file.

On macOS, the YAML-based installation works fine.
On Windows, the environment is created successfully, but the GUI does not launch.

I created the environment from the provided YAML file and activated it, then ran: python -m deeplabcut

DeepLabCut starts loading, but then it falls back to light mode and GUI functionality is unavailable:
Loading DLC 3.0.0rc13...
DLC loaded in light mode; you cannot use any GUI (labeling, relabeling and standalone GUI)

Then I get a Qt-related error: qtpy.QtBindingsNotFoundError: No Qt bindings could be found

I checked the environment, and qtpy, pyside6, pyside6-addons, pyside6-essentials, and qt-main were all present.
However, the following imports fail on Windows:
python -c "from PySide6 import QtCore; print(QtCore.file); print(QtCore.qVersion())"
python -c "from PySide6 import QtWidgets; print(QtWidgets.QApplication)"

ImportError: DLL load failed while importing QtCore: The specified procedure could not be found.
ImportError: DLL load failed while importing QtWidgets: The specified procedure could not be found.

So it seems that:

the package is installed, but the actual Qt DLLs / bindings are not loading correctly on Windows, which causes qtpy to think that no usable Qt binding is available.

-Environment notes
Windows: GUI fails after YAML install
macOS: same YAML-based workflow works fine
GPU is AMD Radeon RX 550, so I am using CPU only
Miniconda is being used

python and conda paths look correct in the environment

Has anyone seen this specific issue on Windows with YAML-based installation?

I’m mainly trying to understand:

  1. whether this is a known PySide6 / Qt DLL issue on Windows,

  2. whether the YAML file pulls a problematic Qt/PySide combination on Windows,

  3. and whether there is a recommended fix for Windows besides manually rebuilding the environment.

Any suggestions would be appreciated.

Activity

  1. reopened this on Mar 22, 2026
  2. deruyter92 commented on Mar 23, 2026

    @deruyter92
    Collaborator

    @RaphJP

    (Edited response)
    This issue seems to be caused by a dll shadowing problem, as a result of mixed conda and pip installations. In this case, adjusting the conda file can likely resolve the problem. I opened PR #3253 to address this.

    Can you try adding pyside6 to the list of conda dependencies in the DEEPLABCUT.yaml file? Like so:

    name: DEEPLABCUT-3
    channels:
      - conda-forge
      - defaults
    dependencies:
      - python=3.10
      - pip
      - ipython
      - jupyter
      - ffmpeg
      - pyside6
      - pip:
        - torch
        - torchvision
        - --pre
        - deeplabcut[gui,modelzoo,wandb]
    
    

    This likely should resolve your problem. Let me know if this helps.

    Note:
    While we currently still support installation via conda yaml for historical reasons, this is not the recommended installation procedure. Since all dependencies are now available independently of conda, we do not need to use mixed conda and pip installations. Below I posted two alternative installation routes that we fully recommend:

    1. Via pip

    create a conda environment (note: this is only the environment, we do not install any packages using conda)

    conda create -n dlc-new python=3.10
    conda activate dlc-new
    

    install pytorch and DeepLabCut

    pip install torch --index-url https://download.pytorch.org/whl/cu126
    pip install --pre deeplabcut[gui,modelzoo,wandb]
    

    run DeepLabCut

    python -m deeplabcut
    

    2. Via uv (faster!)

    Clone the DeepLabCut GitHub repository and move into the created directory

    git clone https://github.com/DeepLabCut/DeepLabCut.git
    cd DeepLabCut
    

    install uv once if you don't have it installed allready.

    pipx install uv
    

    if you are using a GPU with CUDA, on Windows, install pytorch first

    uv pip install torch --index-url https://download.pytorch.org/whl/cu126
    

    install all dependencies using uv

    uv sync --all-extras
    

    run deeplabcut:

    uv run deeplabcut
    
  3. self-assigned this
    on Mar 23, 2026
  4. added theissue type on Mar 24, 2026
  5. RaphJP commented on Mar 25, 2026

    @RaphJP
    Author

    Thanks for the suggestions 👍 :)

    I tried all three methods you suggested, and all of them were successful in launching the GUI on Windows.

    In particular, explicitly adding pyside6 to the YAML dependencies seems to fix the original installation problem, and the environment now installs without the previous Qt/PySide-related error.

    That said, I found one secondary issue that was common across all three methods:
    the GUI only opened successfully when I ran it as Administrator privileges, and in that case the "Choose" button in the "Browse video file" dialog is disabled, so I cannot actually select a video file.

    Running DeepLabCut directly from Jupyter Notebook worked normally, so this seems specific to the GUI file browser.

    So the original installation / GUI startup issue appears to be resolved, but there is still this follow-up issue with file selection under Administrator mode.

    Please let me know if this is a known Windows-specific behavior or if there is a recommended workaround.

  6. C-Achard commented on Mar 25, 2026

    @C-Achard
    Member

    Hello @RaphJP,

    I think the main issue here may be the current default behavior of the “Browse for videos” dialog rather than something inherently Windows-specific, as by default it expects the user to pick a folder of videos, rather than a single video.
    To be allowed to pick a single video, one currently has to check the "Select individual files" checkbox, as shown below.

    Image

    If that checkbox is not enabled, the disabled “Choose” button would be expected behavior when picking a single video file.

    That said, if you still see the same issue even after enabling “Select individual files”, especially only when running as Administrator, that would be useful to know, since Administrator mode itself should not be affecting the UI at all unless something is wrong (it may help with Windows symlinks however).

    Best,
    Cyril

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions