CLI Reference

Main command

Text Only

python -m frd_score <path1> <path2> [OPTIONS]
python -m frd_score --version

path1 and path2 can be:

• Directories containing images (PNG, JPG, TIFF, BMP, NIfTI)
• Paths to .npz statistics files (created with --save_stats)
• Mixed: one .npz and one directory

General flags

Flag	Type	Default	Description
--version	flag	—	Print the installed frd-score version and exit

Extraction & normalisation flags

These flags are shared between the main command and the ood subcommand.

Flag	Type	Default	Description
--frd_version	v0 | v1	v1	FRD version to use
-m, --paths_masks	path path	—	Mask directories for localised extraction
-f, --feature_groups	str...	version default	PyRadiomics feature classes
-I, --image_types	str...	version default	Image filter types: Original, LoG, Wavelet
-r, --resize_size	int [int]	—	Resize images to N×N or W H
-R, --norm_range	float float	version default	Normalisation value range
-T, --norm_type	minmax | zscore	version default	Normalisation strategy
--norm_ref	joint | d1 | independent	version default	Normalisation reference distribution
-v, --verbose	flag	off	Enable detailed logging
-w, --num_workers	int	auto	CPU workers for multiprocessing
--log_sigma	float...	2.0 3.0 4.0 5.0	LoG sigma values
--bin_width	int	5	PyRadiomics bin width
--normalize_scale	float	100	PyRadiomics normalise scale
--voxel_array_shift	float	300	PyRadiomics voxel array shift
--config_path	path	—	Custom PyRadiomics YAML config (overrides above)
--exclude_features	str...	—	Post-extraction exclusion: textural, wavelet, firstorder, shape
--match_sample_count	flag	off	Subsample larger dataset to match smaller

FRD-specific flags

Flag	Type	Default	Description
-s, --save_stats	flag	off	Save statistics to .npz instead of computing FRD
-F, --save_features	flag	off	Save features to CSV
--use_paper_log	flag	off	Use paper Eq. 3 log transform
--means_only	flag	off	Mean-only Fréchet distance (skip covariance)
--interpret	flag	off	Run interpretability analysis
--interpret_dir	path	outputs/interpretability_visualizations	Output dir for interpretation plots

OOD subcommand

Text Only

python -m frd_score ood <reference_path> <test_path> [OPTIONS]

OOD-specific flags

Flag	Type	Default	Description
--detection_type	image | dataset	image	Per-image or dataset-level OOD
--val_frac	float	0.1	Fraction of reference held out for threshold
--use_val_set	flag	off	Enable hold-out validation split
--id_dist_assumption	gaussian | t | counting	gaussian	Statistical model for in-distribution scores
--output_dir	path	outputs/ood_predictions	Output directory for OOD CSV
--seed	int	—	Random seed for reproducibility

All shared extraction flags are also available.

Examples

Bash

# Basic FRDv1
python -m frd_score data/real data/synthetic

# FRDv0 with masks and verbose
python -m frd_score data/real data/synthetic \
    --frd_version v0 \
    -m data/masks_real data/masks_synthetic \
    -v

# Custom feature extraction
python -m frd_score data/real data/synthetic \
    -f firstorder glcm \
    -I Original LoG \
    --log_sigma 1.0 2.0 3.0 \
    --bin_width 10

# Save and reuse statistics
python -m frd_score --save_stats data/reference ref_stats.npz
python -m frd_score ref_stats.npz data/generated

# Exclude shape features (often constant without masks)
python -m frd_score data/real data/synthetic --exclude_features shape

# OOD detection
python -m frd_score ood data/in_domain data/test_images \
    --seed 42 --use_val_set --val_frac 0.2

# Interpretability
python -m frd_score data/real data/synthetic \
    --interpret --interpret_dir results/interp/