Skip to content

v1.6.2: Complete Dependency Declarations & NumPy 2-Safe Floors

Latest

Choose a tag to compare

@rfonod rfonod released this 08 Nov 17:09
4883e02

🐛 v1.6.2: Complete Dependency Declarations & NumPy 2-Safe Floors

A maintenance release that fixes how geo-trax declares its dependencies. No pipeline code changes: outputs, configs, output formats and existing master -> orthophoto caches are unchanged.

A fresh pip install geo-trax already worked; this release makes it keep working, and fixes environments that kept older packages.


📦 Dependencies

Packages geo-trax imports are now declared. numpy, opencv-python, pillow, pyyaml, scipy, shapely, torch and tqdm were imported directly but only arrived as dependencies of other packages. Ultralytics stopped requiring pyyaml after 8.4.184, so it now comes only through stabilo, huggingface_hub and albumentations, yet without it every pipeline command fails at startup (likewise scipy for georeferencing and visualization, and tqdm for almost every stage). Each is now a direct dependency.

Floors that could not work are raised. Several declared minimums predate NumPy 2, and those releases crash next to it:

Package Was Now Why
pandas 2.0 2.2.2 2.0-2.2.1 fail to import next to NumPy 2
seaborn 0.11.0 0.13.0 0.11 does not import with current matplotlib; 0.12 fails the speed-distribution violin plot
matplotlib 3.3.0 3.7.3 older builds fail to import next to NumPy 2 (and 3.3 has no Python 3.11 wheels)
lapx 0.5.2 0.5.10 older builds fail to import next to NumPy 2
shapely (new) 2.0.3 2.0.0-2.0.2 fail to import next to NumPy 2
opencv-python (new) 4.10.0.84 4.10.0.82 fails to import next to NumPy 2
torch (new) 2.3.0 older builds cannot convert NumPy 2 arrays
tifffile (none) 2022.10.10 unbounded
comet_ml (none) 3.40.0 the first with the API Ultralytics' Comet callback calls
  • stabilo>=1.4.8: raises its own opencv-python and torch floors for the same reason; output unchanged.

🧪 Testing

A new test fails whenever the package imports something pyproject.toml does not declare. Verified in fresh environments with every declared minimum installed exactly (resolving to NumPy 1.26.4 and to NumPy 2.3.5) and with the latest versions: the full geotrax batch pipeline on the sample clip, GeoPackage and GeoJSON export, aggregate, and the test suite. 477 tests passing, on Linux, macOS and Windows.


📚 Citation

If you use geo-trax in your research, please cite it (see CITATION.cff or the README citation block, now at v1.6.2).

Full changelog: v1.6.1...v1.6.2