deformaug3d 0.1.1

A package for 3D volume augmentation with configurable parameters

Deformaug3d

Deformaug3d is a Python package for performing 3D volume augmentation with configurable parameters. Designed with simplicity and flexibility in mind, it allows you to augment 3D masks and volumes easily, making it a great choice for machine learning and data augmentation pipelines.

Features

• Configurable Augmentation Parameters: Define your own augmentation ranges or use sensible defaults.
• GPU Support: Leverage CUDA-enabled GPUs for efficient processing via PyTorch.
• Modular Design: Easily integrate with your existing code and extend functionality.

Installation

Requirements

• Python 3.6+
• PyTorch (with CUDA support if available)

Installing from PyPI

(Once published on PyPI, you can install MyPackage using pip)

pip install deformaug3d

Installing Locally

Clone the repository and install in editable mode:

git clone https://github.com/haifangong/deformaug3d.git
cd deformaug3d
pip install -e .

Usage

The package provides functions augmentation and aug_mask_and_img to perform and manage 3D augmentation. Here is an example of how to use these functions:

import torch
from mypackage.augmentation import aug_mask_and_img

# Create dummy data for demonstration
dummy_mask = torch.zeros((1, 1, 128, 128, 128))
dummy_img = torch.ones((1, 1, 128, 128, 128))

# Custom augmentation parameters (optional)
custom_aug_parameters = {
    "rot_range_x": (-10.0, 10.0),
    "rot_range_y": (-8.0, 8.0),
    "rot_range_z": (-8.0, 8.0),
    "scale_range_x": (0.95, 1.00),
    "scale_range_y": (0.95, 1.00),
    "scale_range_z": (0.95, 1.00),
    "shift_range_x": (-0.02, 0.02),
    "shift_range_y": (-0.02, 0.02),
    "shift_range_z": (-0.02, 0.02),
    "contrast": (1.0, 1.0),
    "gray_shift": (0.0, 0.0),
    "flip_x": False,
    "flip_y": False,
    "flip_z": False,
    "elastic_alpha": [3.0, 3.0, 3.0],  # For x, y, z axis
    "smooth_num": 4,
    "field_size": [10, 10, 10],        # For x, y, z axis
    "size_o": (128, 128, 128)
}

# Augment the dummy mask and image using the custom parameters
augmented_mask, augmented_img = aug_mask_and_img(
    dummy_mask, dummy_img, aug_parameters=custom_aug_parameters
)

print("Augmentation complete!")

API Reference

aug_mask_and_img(mask, img, aug_parameters=None)

• mask: Input mask tensor.
• img: Input image/volume tensor.
• aug_parameters (dict, optional): Dictionary of augmentation parameters. If not provided, the default parameters are used.
• Returns: A tuple containing the augmented mask and image/volume as tensors.

augmentation(mask, vol, aug_model)

• mask: Input mask tensor.
• vol: Input volume tensor.
• aug_model: An instance of the augmentation model that applies transformations.
• Returns: A tuple (mask_aug, vol_aug) containing the augmented mask and volume.