axondeepseg / axondeepseg / 10953277948

79%

push

github

`ivadomed` to `nnunetv2` backend migration (#800)

* Add model cards (see issue #799)

* Add SickKids unmyelinated axon seg model

* Switch to pprint for prettier console output

* Move model_cards.py

* Import model cards in download_model.py

* Adapt download_model to use model cards to fetch an nnunet model

* Add argparsing and input validation

* Use loguru for console printing and cleanup old logic

* Change dependency and remove unused stuff

* Add CLI option to list all available models for download

* Add new fn signature for apply_model.axon_segmentation()

* Specify to use python 3.11 for github CI (full disclosure: I am not sure what I am doing)

* Bump miniconda from v2 to v3 for gh CI

* Load nnunetv2 model

* Unpin imageio version

* Add back openpyxl dependency

* Better docstring for get_masks

* Complete axon_segmentation() fn

* Add download_models tests

* apply_model() now fully works for grayscale PNG inputs

* Refactor merge_masks() and update test accordingly

* Support arbitrary model output (e.g. single-class, more than 2, etc.)

* Remove deprecated inference function using ivadomed

* Deprecate unused segment.py arguments (-t, -s, -z, -r, -l, --no-patch and --overlap)

* Refactor main fn and change segment_image/segment_folders signature

* Start refactoring segment_image

* Add get_imshape fn

* Add function to preprocess inputs for nnunet inference

* Silence nnunet when it complains about missing env vars

* Set imageio>=2.28.0, otherwise imread fails on grayscale PNGs

* Apply the model to sanitized inputs in segment_images

* Make prepare_input() cleaner and remove input overwritting during grayscale conversion

* Centralize all suffixes in a tuple for easier import

* Adapt segment_folers()

* Adapt segment_folder tests

* Start adapting segment_images tests

* Adapt get_existing_models test

* Try installing numpy through pip instead of conda
... (continued)

213 of 218 new or added lines in 10 files covered. (97.71%)

1046 of 1106 relevant lines covered (94.58%)

2.84 hits per line