cli.md

🛠️ CLI Reference

AnomaVision provides a unified anomavision command with four subcommands:

anomavision train    # Train a PaDiM anomaly detection model
anomavision detect   # Run inference on test images
anomavision eval     # Evaluate performance on MVTec-style datasets
anomavision export   # Export models to ONNX, TorchScript, or OpenVINO

Each subcommand accepts both CLI arguments and config files (--config config.yml). CLI arguments always override config file values.

Get help at any level:

anomavision --help
anomavision train --help
anomavision detect --help
anomavision eval --help
anomavision export --help

---

1. Training — anomavision train

anomavision train [options]

Argument	Type	Default	Description
--config	str	config.yml	Path to config file
--dataset_path	str	None	Dataset root containing train/good
--resize	int(s)	None	Resize images (one value = square, two values = W H)
--crop_size	int(s)	None	Crop size (one = square, two = W H)
--normalize	flag	False	Enable normalization
--no_normalize	flag	False	Disable normalization (overrides --normalize)
--norm_mean	float(3)	None	Normalization mean (RGB)
--norm_std	float(3)	None	Normalization std (RGB)
--backbone	str	resnet18	Feature extractor (resnet18, wide_resnet50)
--batch_size	int	2	Batch size
--feat_dim	int	50	Number of random features
--layer_indices	int list	[0]	Backbone layer indices
--output_model	str	model.pt	Model filename (.pt)
--run_name	str	anomav_exp	Experiment name
--model_data_path	str	./distributions	Output directory
--log_level	str	INFO	Logging level

Example:

anomavision train \
  --config config.yml \
  --dataset_path ./dataset \
  --backbone resnet18 \
  --batch_size 16

---

2. Detection — anomavision detect

anomavision detect [options]

Argument	Type	Default	Description
--config	str	None	Path to config file
--img_path	str	None	Path to test images
--model_data_path	str	./distributions/anomav_exp	Directory with model files
--model	str	model.pt	Model file (.pt, .onnx, .engine)
--device	str	auto	Device (cpu, cuda, auto)
--batch_size	int	1	Batch size
--thresh	float	None	Anomaly threshold
--num_workers	int	1	Data loader workers
--pin_memory	flag	False	Use pinned memory (GPU transfer)
--enable_visualization	flag	False	Show anomaly maps
--save_visualizations	flag	False	Save images to disk
--viz_output_dir	str	./visualizations	Save path
--run_name	str	detect_exp	Experiment name
--overwrite	flag	False	Overwrite run dir
--log_level	str	INFO	Logging level
--detailed_timing	flag	False	Log detailed timings

Example:

anomavision detect \
  --config config.yml \
  --img_path ./dataset/bottle/test \
  --thresh 13.0 \
  --enable_visualization \
  --save_visualizations

---

3. Evaluation — anomavision eval

anomavision eval [options]

Argument	Type	Default	Description
--config	str	None	Path to config file
--dataset_path	str	None	Root dataset path
--class_name	str	bottle	Class name (MVTec style)
--model_data_path	str	./distributions/anomav_exp	Directory with model files
--model	str	model.onnx	Model file
--device	str	auto	Device (cpu, cuda)
--batch_size	int	32	Batch size
--num_workers	int	1	Data loader workers
--pin_memory	flag	False	Use pinned memory
--enable_visualization	flag	False	Show plots
--save_visualizations	flag	False	Save plots
--viz_output_dir	str	./eval_visualizations	Output path
--log_level	str	INFO	Logging level
--detailed_timing	flag	False	Log detailed timings

Example:

anomavision eval \
  --config config.yml \
  --dataset_path ./dataset \
  --class_name bottle \
  --enable_visualization

---

4. Export — anomavision export

anomavision export [options]

Argument	Type	Default	Description
--config	str	None	Path to config file
--model_data_path	str	./distributions/anomav_exp	Directory with model & outputs
--model	str	(required)	Model file (.pt)
--format	str	(required)	Export format (onnx, torchscript, openvino, all)
--device	str	auto	Export device
--precision	str	auto	Precision (fp32, fp16, auto)
--opset	int	17	ONNX opset version
--static-batch	flag	False	Disable dynamic batch
--optimize	flag	False	Optimize TorchScript for mobile
--quantize-dynamic	flag	False	Export dynamic INT8 ONNX
--quantize-static	flag	False	Export static INT8 ONNX (needs calibration)
--calib-samples	int	100	Calibration samples
--log_level	str	INFO	Logging level

Example:

anomavision export \
  --model_data_path ./distributions/anomav_exp \
  --model model.pt \
  --format onnx \
  --precision fp16 \
  --quantize-dynamic

---

Config File + CLI Override Pattern

All subcommands follow the same priority: CLI args > config file > defaults.

# Config sets backbone=resnet18, CLI overrides to wide_resnet50
anomavision train --config config.yml --backbone wide_resnet50

This makes it easy to run sweeps or one-off experiments without editing config files.