Skip to content

Conversation

@dyollb
Copy link

@dyollb dyollb commented Aug 8, 2025

Partially fixes issue #2396

fix windows ci

normalize gaussian using numpy

  • as noted here, normalization using torch produces 'Inf' for the test case with torch==2.4.0.
  • doing the operation on the numpy array instead should not have any negative side-effects

specify folder 'nnunetv2' in pyproject.toml

  • doesn't hurt. I noticed when I had temporary folders (e.g., 'venv/'), setuptools was not able to install the package properly
  • add .venv/ to the gitignore file. This is a common name for a virtual environment

note: this PR does not 'fix' the issue with 2.4.0. I tried accumulating the predicted_logits in predict_from_raw_data.py using float32, which also helped. But in the end I still got a dice of 0.008

@FabianIsensee FabianIsensee self-assigned this Oct 16, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants