nndet_compute_metric_results script#
Script to perform Object Detection (COCO and FROC Metrics) and Segmentation (Dice score) evaluation on nnDetection experiments.
By specifying the class-file and classes parameters, the class-wise analysis of the metrics is performed.
An Excel Spreadsheet and several PNG plots (representing FROC curves and Histogram Analysis), are produced as output.
usage: nndet_compute_metric_results [-h] --config-file CONFIG_FILE
[--run-fold RUN_FOLD] --output-dir
OUTPUT_DIR [--class-file CLASS_FILE]
[--classes CLASSES [CLASSES ...]]
[--n-folds N_FOLDS] [--model MODEL]
[-v | -q]
Named Arguments#
- --config-file
File path for the configuration dictionary, used to retrieve experiments variables.
- --run-fold
Index indicating which fold to run. Default:
-1. If set to-1, runs the metric evaluation on the consolidated predictions.Default:
'-1'- --output-dir
Folder path where to save the Excel spreadsheet and the PNG plots containing the evaluation metrics.
- --class-file
Optional JSON file, containing a dictionary where each Subject ID is stored with the corresponding Subject Class.If not specified, all the subjects will be analysed.
- --classes
Optional list of Subject Classes, to be considered for the class-wise result analysis.
- --n-folds
Number of Folds used to aggregate subjects and create Object Detection metrics statistic.
Default:
'1'- --model
nnDetection Model used for sweeping and consolidation. Default:
RetinaUNetV001_D3V001_3dDefault:
'RetinaUNetV001_D3V001_3d'- -v, --verbose
Enable verbose output in terminal. Add multiple times to increase verbosity.
- -q, --silent
Suppress most log outputs in terminal.
Example call:
nndet_compute_metric_results --config-file /PATH/TO/CONFIG_FILE.json --output-dir /OUTPUT/PATH
nndet_compute_metric_results --config-file /PATH/TO/CONFIG_FILE.json --output-dir /OUTPUT/PATH --n-fold 0