Skip to content

Troubleshooting

Troubleshooting

This page covers problems common to all versions of ASHS. Each version also has its own troubleshooting page:

Cloud Local
HyperASHS Troubleshooting Cloud HyperASHS Troubleshooting HyperASHS
ASHS Classic Troubleshooting Cloud ASHS Troubleshooting Local ASHS

SSL certificate errors when downloading models

hrashs downloads models and templates from Hugging Face. On some institutional or corporate networks, a proxy or firewall intercepts secure connections, and the download fails with an error that mentions SSL or certificate verification (for example, SSLCertVerificationError or CERTIFICATE_VERIFY_FAILED). hrashs check reports this as:

SSL certificate verification failed when trying to connect to Hugging Face Hub.

To work around this, add the -k (--disable-ssl-verification) option to the hrashs command. It works with run, list, desc, check, train and upload:

hrashs check -k
hrashs run -k -a PMC_Atlas -g sub01_mprage.nii.gz -f sub01_tse.nii.gz -I sub01 -w ./output/sub01

Warning

-k turns off the check that you are really connected to Hugging Face. Use it only on a network you trust. The better long-term fix is to ask your IT department for your institution's root certificate and point Python to it, for example with export REQUESTS_CA_BUNDLE=/path/to/institution-ca.pem.

Results look wrong

Most problems with segmentation quality come from the input data or from registration:

  • Check that your MRI data meet the requirements of the model or atlas you are using: see MRI Data for HyperASHS or ASHS Classic.
  • Look at the registration QC images. If the T1-weighted and T2-weighted scans are not aligned, register them manually in ITK-SNAP.
  • When measuring volumes with HyperASHS, use the hyper-resolution segmentation rather than the native-space one (see Outputs).

Getting help

If none of the above helps, ask on the ashs-users mailing list or open an issue on GitHub. Please include:

  • the version of ASHS you are using (HyperASHS or ASHS Classic, cloud or local), and the model or atlas;
  • the exact command you ran and the error message, or the ticket number for cloud runs;
  • the output of hrashs check (HyperASHS) and the QC images, if relevant.