CLI Reference

DeepFense provides a CLI for training, testing, and inspecting components.

Setup

pip install -e .       # installs the `deepfense` command
deepfense --help       # verify it works

Commands

deepfense train

Train a model from a YAML config.

deepfense train --config deepfense/config/train.yaml
deepfense train -c deepfense/config/train.yaml --resume outputs/exp/best_model.pth
Option Short Required Description
--config -c Yes Path to YAML config file
--resume -r No Resume from a checkpoint

deepfense test

Evaluate a trained model on the test set defined in the config.

deepfense test --config deepfense/config/train.yaml --checkpoint outputs/exp/best_model.pth
Option Short Required Description
--config -c Yes Path to YAML config file
--checkpoint -ckpt Yes Path to .pth checkpoint

deepfense list

Show all registered components (frontends, backends, losses, etc.).

deepfense list                           # show everything
deepfense list --component-type backends # show only backends
Option Short Default Choices
--component-type -t all all, frontends, backends, losses, datasets, augmentations, optimizers, trainers

deepfense download

Download datasets and pretrained models from HuggingFace.

# List available datasets
deepfense download list-datasets

# List available models (with optional filter)
deepfense download list-models
deepfense download list-models --filter WavLM --limit 50

# Download a dataset
deepfense download dataset CompSpoof
deepfense download dataset ASVSpoof19 --output-dir ./my_data

# Download a pretrained model
deepfense download model ASV19_WavLM_Nes2Net_NoAug_Seed42
deepfense download model ASV19_WavLM_Nes2Net_NoAug_Seed42 --output-dir ./my_models

See the HuggingFace Hub Guide for full usage details.

Equivalent Python Scripts

The CLI wraps the same logic as the standalone scripts:

CLI Script
deepfense train -c config.yaml python train.py --config config.yaml
deepfense test -c config.yaml -ckpt model.pth python test.py --config config.yaml --checkpoint model.pth

Both work identically. Use whichever you prefer.