I’m trying to use the TotalSegmentator extension in 3D Slicer for segmenting an MRI dataset, but I’m running into a consistent error during the dependency installation step.
When clicking “Apply” in the TotalSegmentator module, I get:
PyTorch Python package is required. Installing… (it may take several minutes)
Failed to install Python dependencies:
PyTorchUtilsLogic.installTorch() got an unexpected keyword argument ‘torchVisionVersionRequirement’
nvironment:
OS: Windows (also tested on macOS with same error)
Slicer version: 5.11.0 (Preview)
(Also tested on 5.10 stable with same issue
Extensions installed:
TotalSegmentator (latest)
PyTorch (latest via Extensions Manager)
nnUNet (latest)
Observations: PyTorch appears installed and visible in Extensions Manager. Error occurs during the automatic dependency installation step triggered by TotalSegmentator. Suggests a mismatch between TotalSegmentator and PyTorchUtils API (unexpected argument)
Is this a known compatibility issue between TotalSegmentator and current PyTorchUtils versions? Is there a recommended Slicer version / extension version combination that avoids this? Is there a way to bypass the internal PyTorch installation step and point to an existing PyTorch install?
Sorry, that’s due to a recent regression that I cause when fixing the macOS installation. I’ve fixed the issue now, please retry tomorrow. You can either use the latest Slicer Stable Release (Slicer-5.10) and update the TotalSegmentator extension; or install the latest Slicer Preview Release (Slicer-5.11) tomorrow.
You can make these changes locally in TotalSegmentator.py file on your computer (you can find the location of the file in the error message in the Python console); or wait for the nightly builds to complete at around 10am EST.
Great, the precoous problem was solved. TotalSegmentator is now started.
Do you see more specific error message in the application log? What GPU do you have? Is segmentation successful on the CTChest sample sata set in fastest mode? Is segmentation successful if you force to use CPU?