Skip to content

v0.26.0

Compare
Choose a tag to compare
@Pale-Blue-Dot-97 Pale-Blue-Dot-97 released this 12 Sep 13:46
· 1507 commits to main since this release
8ab77fe

v0.26.0

This release adds new types of models, specialist support for new datasets, auto-normalisation and quality-of-life and bug fixes.

Note

Due to unknown reasons, the CircleCI workflow and Qodana scan actions are failing despite the tests passing. Be assured, this release is still tested to the same standards as previous releases.

Major Updates ⬆

  • ✨ NEW: Convolutionally based Siamese model, SimConv
  • ✨ NEW: SegBarlowTwinsLoss, a Barlow Twins loss adapted for use with 2D outputs
  • ✨ NEW: Adaption of a PSP for use as an encoder PSPEncoder
  • ✨ NEW: Dataset class for Sentinel2 data that comes with the SSL4EO-S12 dataset
  • ✨ NEW: Added a notebook to demonstrate using the SSL4EO-S12 dataset within Minerva
  • ✨ NEW: You can now use the AutoNorm transform to automatically normalise a dataset by its mean and std characteristics found in its metadata. This usage is integrated into make_datasets
  • ✨ NEW: DFC2020 dataset class that works with torchgeo
  • 🛠 Reorganised the datasets.py module into a sub-package
  • 🛠 Reorganised the test suite to match the package structure where possible
  • 🟢 Can now parse the CRS in dataset params in yaml and a CRS will be created and parsed to the dataset on construction.

Bug Fixes 🐞

  • 🐞 Fixed bugs with accessing the model wrapped up in a MinervaDistributed object

Other Changes 🔁

  • 🛑 Dropped support for Python 3.8
  • ⬆ Various dependency upgrades
  • ⬆ Various actions and hook updates

Changes

Full Changelog: v0.25.0...v0.26.0

Contributors