evaluate-segmentations DAG
closes #798 (closed)
Adds a new segmentation evaluation DAG that takes a single dataset as input, and separates it into two sets of segmentations (two separate folders) based on a tag. For instance, a tag like "pred" to indicate that these segmentations are from prediction set, and the rest are from ground truth. Then it performs filtering, fusing based on labels, and finally perfoms the evaluation.
Changes include:
- new segmentation evaluation DAG that compares two sets of segmentations, GT and test
- new SegmentationEvaluationOperator
- for FL: new (experimental) prediction evaluation DAG that runs an nnunet_predict on GT and afterwards compares GT and test
- HelperOpenSearch:
include_custom_tag
andexclude_custom_tag
logic - LocalGetInputData:
include_custom_tag_property
andexclude_custom_tag_property
- LocalFilterMasks: use
label_filter_key
parameter instead of hard codedlabel_filter
to work with multiple filters in ui_forms - MergeMasks: use
fuse_labels_key
andfused_label_name_key
to specify keys from ui_forms. Also add logic to fuse multiple masks with same label