Skip to content

0.3.0

Latest
Latest

Choose a tag to compare

View all tags
@simonreise simonreise released this 26 Mar 17:20
· 21 commits to main since this release
v0.3.0
93fb7d5

ML module changes

  • Now semantic and regression are separate modules
  • Added models from segmentation_models_pytorch and torchgeo libraries
  • Added more sklearn models
  • Completely changed the syntax
  • Code base is almost completely rewritten
  • ML datasets are now stored in .rspds format - a custom ML dataset format based on Huggingface Datasets library
  • Now any custom Pytorch or Sklearn-based model can be trained
  • More metrics available, including custom metrics
  • Loss selection available, including custom losses
  • Now user can define the augmentations that will be applied to train dataset, including custom augmentations
  • Now basic multiscale features can be generated to improve the modeling quality of Sklearn-based models
  • Added band_importance functions that use SHAP to estimate band importance for the modeling
  • Added confusion_matrix function that calculate confusion matrix for semantic segmentation models

Other major changes

  • Added process function that can clip, reproject, reproject match and change dtype of a single raster
  • Added dem group of functions: slope, curvature, aspect, hillshade
  • Added match_hist function that matches histograms of two images/datasets
  • Added clip_values function that clips raster values to a certain range
  • Added denormalize functions that restore original values from normalized data
  • Added get_normalization_params functions that retrieve optimal normalization parameters
  • Added zscore and dynamicworld normalization
  • Now data is saved with a metadata file in STAC format (can be controlled with write_stac argument)
  • Almost every function now supports not only file paths, but also STAC Items as inputs
  • Most of the functions now can process multi-band datasets and STAC datasets
  • Reworked calculate_index function, now it supports all the indices supported by spyndex library

Minor changes

  • replace now supports multiple values replacement via values arg
  • Added clip_values and nodata args to normalize function
  • input_file and output_file args are renamed to input_path and output_path
  • process now accepts dtype arg, which will convert input dataset to the requested dtype
  • Landsat imagery is now processed by satpy
  • Sentinel-2 superresolution models are now stored on Huggingface Hub
  • sen2cor parameter of sentinel2 function is now False. Sen2Cor support is going to be deprecated in the future.
  • The required Sen2Cor version is now 02.12.03
  • Multiple performance optimisations

Deprecations

  • landsat no longer supports Collection-1 products because they are no longer available to download
Assets 2
Loading