Pipeline API Reference

Technical reference for the pipeline module.

Module: pipeline.runner

run_pipeline(config)

Main entry point for the autoStructN2V pipeline.

Parameters:

• config (dict): Configuration dictionary

Returns:

• dict: Results summary with keys:
  • experiment_dir: Path to experiment directory
  • config: Complete configuration
  • stages_run: List of stages executed
  • stage1_model_path: Path to Stage 1 model (if run)
  • stage1_denoised_dir: Path to Stage 1 outputs (if run)
  • stage2_model_path: Path to Stage 2 model (if run)
  • stage2_mask_path: Path to Stage 2 mask (if run)
  • final_results_dir: Path to final denoised images

Raises:

• ValueError: Invalid configuration
• FileNotFoundError: Input directory not found
• RuntimeError: Training or inference failures

Example:

from autoStructN2V.pipeline import run_pipeline

config = {
    'input_dir': './data/',
    'experiment_name': 'my_exp',
}
results = run_pipeline(config)

---

create_stage2_mask(config, image_paths, denoised_patches, verbose)

Create structured mask for Stage 2.

Parameters:

• config (dict): Configuration
• image_paths (tuple, optional): (train, val, test) paths
• denoised_patches (np.ndarray, optional): Stage 1 patches
• verbose (bool): Print details

Returns:

• tuple: (full_mask, prediction_kernel)
  • full_mask: Boolean array (patch_size, patch_size)
  • prediction_kernel: Boolean array (patch_size, patch_size)

Example:

full_mask, pred_kernel = create_stage2_mask(
    config, image_paths=paths, verbose=True
)

---

load_mask_from_file(mask_file_path, verbose)

Load masking kernel from .npy file.

Parameters:

• mask_file_path (str): Path to .npy file
• verbose (bool): Print loading details

Returns:

• np.ndarray: Boolean mask array

Raises:

• FileNotFoundError: File doesn't exist
• ValueError: Invalid mask format

---

denoise_directory(model, input_dir, output_dir, config, stage)

Apply trained model to directory.

Parameters:

• model (nn.Module): Trained model
• input_dir (str): Input directory
• output_dir (str): Output directory
• config (dict): Configuration
• stage (str): 'stage1' or 'stage2'

Returns:

• list: Paths to denoised images

---

Module: pipeline.config

validate_config(config)

Validate and complete configuration.

Parameters:

• config (dict): User configuration

Returns:

• dict: Validated configuration with defaults

Raises:

• ValueError: Invalid configuration

Validation Checks:

• Required fields present
• Stage execution flags valid
• Mask source configuration correct
• Numeric parameters in valid ranges
• Paths exist (for file-based masks)

---

create_output_directories(config)

Create output directory structure.

Parameters:

• config (dict): Validated configuration

Returns:

• dict: Dictionary of directory paths

Created Structure:

output_dir/experiment_name/
├── data/
│   ├── train/
│   ├── val/
│   └── test/
├── stage1/ (if enabled)
│   ├── model/
│   ├── logs/
│   └── results/
├── stage2/ (if enabled)
│   ├── model/
│   ├── logs/
│   └── results/
└── final_results/

---

Module: pipeline.data

split_dataset(input_dir, output_dirs, split_ratio, image_extension, seed, verbose)

Split images into train/val/test sets.

Parameters:

• input_dir (str): Input directory
• output_dirs (dict): Output directory structure
• split_ratio (tuple): (train, val, test) ratios
• image_extension (str): Image file extension
• seed (int): Random seed
• verbose (bool): Print split details

Returns:

• tuple: (train_paths, val_paths, test_paths)

Example:

paths = split_dataset(
    './data/',
    dirs,
    (0.7, 0.15, 0.15),
    '.tif',
    42,
    verbose=True
)

---

create_dataloaders(image_paths, config, stage, stage1_denoised_dir, structured_mask, prediction_kernel, verbose)

Create PyTorch DataLoaders.

Parameters:

• image_paths (tuple): (train, val, test) paths
• config (dict): Configuration
• stage (str): 'stage1' or 'stage2'
• stage1_denoised_dir (str, optional): Stage 1 output dir
• structured_mask (np.ndarray, optional): Structured mask
• prediction_kernel (np.ndarray, optional): Prediction kernel
• verbose (bool): Print loader details

Returns:

• tuple: (train_loader, val_loader, test_loader)

Example:

train_loader, val_loader, test_loader = create_dataloaders(
    paths, config, 'stage1', verbose=True
)

---

See Also

• Pipeline Guide - Usage guide
• Configuration Reference - All parameters
• Architecture - System design