API Reference

This page documents all the exported functions in py-distance-transforms.

Core Functions

transform

transform(arr)

Computes the squared Euclidean distance transform of a binary array.

Parameters: - arr : numpy.ndarray - A binary array (0s and 1s) of any dimension

Returns: - numpy.ndarray - The squared Euclidean distance transform

Example:

import numpy as np
from py_distance_transforms import transform

arr = np.array([
    [0, 1, 1, 0, 1],
    [0, 0, 0, 1, 0],
    [1, 1, 0, 0, 0]
], dtype=np.float32)

result = transform(arr)
print(result)

transform_cuda

transform_cuda(tensor)

GPU-accelerated version of the distance transform function.

Parameters: - tensor : torch.Tensor - A binary tensor (0s and 1s) of any dimension, located on a CUDA device

Returns: - torch.Tensor - The squared Euclidean distance transform, located on the same CUDA device

Example:

import torch
from py_distance_transforms import transform_cuda

# Create a tensor on GPU
tensor = torch.zeros((10, 10), device='cuda')
tensor[2, 2] = 1
tensor[7, 7] = 1

result = transform_cuda(tensor)
print(result)

Utility Functions

numpy_to_torch

numpy_to_torch(arr, device='cuda')

Converts a NumPy array to a PyTorch tensor on the specified device.

Parameters: - arr : numpy.ndarray - Input NumPy array - device : str, optional - Device to place the tensor on. Default is ‘cuda’

Returns: - torch.Tensor - PyTorch tensor on the specified device

torch_to_numpy

torch_to_numpy(tensor)

Converts a PyTorch tensor to a NumPy array.

Parameters: - tensor : torch.Tensor - Input PyTorch tensor

Returns: - numpy.ndarray - NumPy array

Full Function List

Function	Description
`transform`	Computes squared Euclidean distance transform on CPU
`transform_cuda`	Computes squared Euclidean distance transform on GPU
`numpy_to_torch`	Converts NumPy array to PyTorch tensor
`torch_to_numpy`	Converts PyTorch tensor to NumPy array

Integration with Other Libraries

NumPy

py-distance-transforms works seamlessly with NumPy arrays:

import numpy as np
from py_distance_transforms import transform

# Create a NumPy array
arr = np.zeros((100, 100), dtype=np.float32)
arr[25:75, 25:75] = 1

# Apply transform
result = transform(arr)

PyTorch

For PyTorch integration, use the CUDA-specific functions:

import torch
from py_distance_transforms import transform_cuda

# Create a PyTorch tensor
tensor = torch.zeros((100, 100), device='cuda')
tensor[25:75, 25:75] = 1

# Apply transform
result = transform_cuda(tensor)

scikit-image

Works alongside scikit-image for more advanced image processing:

import numpy as np
from skimage import io, color, morphology
from py_distance_transforms import transform

# Load and process image
img = io.imread('image.png', as_gray=True)
binary = img > 0.5

# Apply morphological operations
cleaned = morphology.remove_small_objects(binary)

# Apply distance transform
dist = transform(cleaned.astype(np.float32))

--- title: "API Reference" sidebar: python format: html: toc: true --- This page documents all the exported functions in py-distance-transforms. ## Core Functions ### transform ```python transform(arr) ``` Computes the squared Euclidean distance transform of a binary array. **Parameters:** - `arr` : numpy.ndarray - A binary array (0s and 1s) of any dimension **Returns:** - numpy.ndarray - The squared Euclidean distance transform **Example:** ```python import numpy as np from py_distance_transforms import transform arr = np.array([ [0, 1, 1, 0, 1], [0, 0, 0, 1, 0], [1, 1, 0, 0, 0] ], dtype=np.float32) result = transform(arr) print(result) ``` ### transform_cuda ```python transform_cuda(tensor) ``` GPU-accelerated version of the distance transform function. **Parameters:** - `tensor` : torch.Tensor - A binary tensor (0s and 1s) of any dimension, located on a CUDA device **Returns:** - torch.Tensor - The squared Euclidean distance transform, located on the same CUDA device **Example:** ```python import torch from py_distance_transforms import transform_cuda # Create a tensor on GPU tensor = torch.zeros((10, 10), device='cuda') tensor[2, 2] = 1 tensor[7, 7] = 1 result = transform_cuda(tensor) print(result) ``` ## Utility Functions ### numpy_to_torch ```python numpy_to_torch(arr, device='cuda') ``` Converts a NumPy array to a PyTorch tensor on the specified device. **Parameters:** - `arr` : numpy.ndarray - Input NumPy array - `device` : str, optional - Device to place the tensor on. Default is 'cuda' **Returns:** - torch.Tensor - PyTorch tensor on the specified device ### torch_to_numpy ```python torch_to_numpy(tensor) ``` Converts a PyTorch tensor to a NumPy array. **Parameters:** - `tensor` : torch.Tensor - Input PyTorch tensor **Returns:** - numpy.ndarray - NumPy array ## Full Function List | Function | Description | |----------|-------------| | `transform` | Computes squared Euclidean distance transform on CPU | | `transform_cuda` | Computes squared Euclidean distance transform on GPU | | `numpy_to_torch` | Converts NumPy array to PyTorch tensor | | `torch_to_numpy` | Converts PyTorch tensor to NumPy array | ## Integration with Other Libraries ### NumPy py-distance-transforms works seamlessly with NumPy arrays: ```python import numpy as np from py_distance_transforms import transform # Create a NumPy array arr = np.zeros((100, 100), dtype=np.float32) arr[25:75, 25:75] = 1 # Apply transform result = transform(arr) ``` ### PyTorch For PyTorch integration, use the CUDA-specific functions: ```python import torch from py_distance_transforms import transform_cuda # Create a PyTorch tensor tensor = torch.zeros((100, 100), device='cuda') tensor[25:75, 25:75] = 1 # Apply transform result = transform_cuda(tensor) ``` ### scikit-image Works alongside scikit-image for more advanced image processing: ```python import numpy as np from skimage import io, color, morphology from py_distance_transforms import transform # Load and process image img = io.imread('image.png', as_gray=True) binary = img > 0.5 # Apply morphological operations cleaned = morphology.remove_small_objects(binary) # Apply distance transform dist = transform(cleaned.astype(np.float32)) ```