Error running the sample

[DrMaxP]

I've followed the steps to set up the environment, installing as a package. When I go to run the sample I see errors:
e..g running this:

priorda test --image_path assets/sample-1/rgb.jpg --prior_path assets/sample-1/gt_depth.png --pattern downscale_32 --visualize 1

Firstly, the message xFormers not available is printed out twice.

I then see the process terminate with this call stack:

 C:\Users\Max\anaconda3\envs\priorda\Lib\site-packages\torch_cluster\_version_cpu.pyd
Traceback (most recent call last):
  File "\\?\C:\Users\Max\anaconda3\envs\priorda\Scripts\priorda-script.py", line 33, in <module>
    sys.exit(load_entry_point('prior-depth-anything', 'console_scripts', 'priorda')())
  File "\\?\C:\Users\Max\anaconda3\envs\priorda\Scripts\priorda-script.py", line 25, in importlib_load_entry_point
    return next(matches).load()
  File "C:\Users\Max\anaconda3\envs\priorda\lib\importlib\metadata.py", line 86, in load
    module = import_module(match.group('module'))
  File "C:\Users\Max\anaconda3\envs\priorda\lib\importlib\__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 972, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 986, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 680, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 850, in exec_module
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "c:\users\max\documents\kinect experiments\depth priors\prior-depth-anything\prior_depth_anything\__init__.py", line 14, in <module>
    from .depth_completion import DepthCompletion
  File "c:\users\max\documents\kinect experiments\depth priors\prior-depth-anything\prior_depth_anything\depth_completion.py", line 3, in <module>
    import torch_cluster
  File "C:\Users\Max\anaconda3\envs\priorda\lib\site-packages\torch_cluster\__init__.py", line 18, in <module>
    torch.ops.load_library(spec.origin)
  File "C:\Users\Max\anaconda3\envs\priorda\lib\site-packages\torch\_ops.py", line 933, in load_library
    ctypes.CDLL(path)
  File "C:\Users\Max\anaconda3\envs\priorda\lib\ctypes\__init__.py", line 376, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: [WinError 127] The specified procedure could not be foundanother error:

Any thoughts? I'm on Windows with a 4090.

I've tried installing xformers via pip, but that started pulling in conflicting package versions.

[DrMaxP]

Output from pip list:

Package              Version     Editable project location
-------------------- ----------- ---------------------------------------------------------------------------
certifi              2025.6.15
charset-normalizer   3.4.2
colorama             0.4.6
contourpy            1.3.0
cycler               0.12.1
einops               0.8.1
filelock             3.18.0
fonttools            4.58.4
fsspec               2025.5.1
hloc                 1.5
huggingface-hub      0.26.5
idna                 3.10
importlib_resources  6.5.2
Jinja2               3.1.6
kiwisolver           1.4.7
MarkupSafe           3.0.2
matplotlib           3.9.3
mpmath               1.3.0
networkx             3.2.1
numpy                1.25.2
opencv-python        4.10.0.84
packaging            25.0
pillow               11.0.0
pip                  25.1.1
prior_depth_anything 1.0         c:\users\max\documents\kinect experiments\depth priors\prior-depth-anything
pyparsing            3.2.3
python-dateutil      2.9.0.post0
PyYAML               6.0.2
requests             2.32.4
safetensors          0.4.5
scipy                1.13.1
setuptools           78.1.1
six                  1.17.0
sympy                1.14.0
torch                2.2.2
torch_cluster        1.6.3
torchvision          0.17.2
tqdm                 4.67.1
typing_extensions    4.14.0
urllib3              2.5.0
wheel                0.45.1
zipp                 3.23.0

[DrMaxP]

Error dialog:

[DrMaxP]

I've gone through the setup procedure twice, wiping the original environment. Still no joy.

[Ammoniro]

Hello, in my experience, "xFormer" is not necessary. Our implementation is based on Ubuntu 22.04.

[MaxPalmer-UH]

OK, I presume that means this doesn't work on Windows. Can anyone confirm this please?

[MaxPalmer-UH]

Also @Ammoniro which CUDA version are you using / would you recommend?

[DrMaxP]

Looks like the issue is loading the torch_cluster _version_cpu.pyd file.

If you drill down into the dependencies there are 3 dlls it can't find. I can find almost no info on these and where I'd expect to find/obtain them.

Also, I'm confused. I'm installing the cuda version of the package, but the torch_cluster pyd files are all the _cpu variants, not the _cuda variants. I have various cuda versions installed, the active one is 12.1 which I believe matches what's used here.

[DrMaxP]

OK, from this post https://learn.microsoft.com/en-us/answers/questions/1195445/kernel32-dll-dependenci-ext-ms-win-oobe-query-l1-1 it seems like any dlls starting with ext represent APIs that may not exist and the system will not attempt to resolve loading these at load time. So the weird thing is that several dependency checking / walking tools are otherwise happy to load the torch_cluster pyd file. However, the python code is not happy. Why?

[DrMaxP]

Am I alone in getting this error?

I've cleaned out all python packages across my system and gone through the installation steps several times. I always hit this error (see the thread) - the _version_cpu.pyd file that I get does not contain a method that's expected, I believe it's in c10.dll which is a dependency. The method is something like DeviceGuardImplRegistry, which would be consistent with this interface https://github.com/pytorch/pytorch/blob/main/c10/core/impl/DeviceGuardImplInterface.h.

I also don't understand why my system is not using the CUDA versions of the pyd files but always the cpu versions. I have CUDA set up, it's in the PATH, the system variables point to the correct version which is used here 12.1. I've tried fetching the cuda variant of torch_cluster. Note other python projects are using CUDA, so I'm stumped.

Note: the error is very similar to this one which is why I cleared out all packages including in the (base) https://stackoverflow.com/questions/73931064/how-to-fix-entry-point-not-found-while-import-package-from-torch-sparse

[DrMaxP]

So I've finally got the sample to run by somehow forcing the cuda version of the torch packages to be installed - I switched from pip to conda for package management. Note, by default the output path parent = datetime.now().strftime("%Y-%m-%d %H:%M") does not work on Windows as : is an illegal filename character. I now have a different issue running on my data, but the sample does work.