Euclidean Distance Transform 3D

3.1.1 verified Fri May 01 auth: no python

Computes the Euclidean distance transform of 3D and 2D label fields. Supports anisotropic voxels (non-cubic) and multi-label segmentation. v3.1.1, updated June 2024, active development with irregular releases.

pip install edt

Common errors

error ImportError: cannot import name 'edt' from 'edt' ↓

cause Trying to import edt as from edt import edt, but edt is the module itself.

fix

Use just 'import edt' then call 'edt.edt()'.

error ValueError: ndim is 2 but expected 3 ↓

cause Passed a 2D array to edt.edt without specifying ndim=2 or using the 2D variant.

fix

For 2D arrays, use edt.edt(..., ndim=2) or call edt.edt2d() if available (not in current version).

error TypeError: anisotropic must be a tuple of floats ↓

cause Anisotropic parameter passed as list or wrong type.

fix

Pass anisotropic as a tuple of three floats, e.g., (0.5, 0.5, 1.0).

Warnings

gotcha edt.edt expects uint32 or uint64 labels by default. Passing int32 or float labels will silently produce incorrect results. ↓

fix Convert labels to np.uint32 or np.uint64 before calling edt.edt.

gotcha Anisotropic parameter must be a tuple of three floats (for 3D) or two floats (for 2D). Omitting or using wrong shape leads to undefined behavior. ↓

fix Always provide anisotropic as (sx, sy, sz) tuple for 3D, or (sx, sy) for 2D.

gotcha edt.edt modifies the input array in-place if the 'in_place' parameter is True (default False). Unexpected side effects may occur. ↓

fix If you need to preserve the original array, set in_place=False (the default) or pass a copy.

deprecated The 'parallel' parameter (int) was deprecated in 3.0.0 and removed in 3.1.0. Use 'num_threads' instead. ↓

fix Use num_threads=N instead of parallel=N.

Imports

edt
wrong
```
from edt import edt
```
correct
```
import edt
```
edt is a module, not a function within edt; top-level import is correct.

Quickstart

Compute Euclidean distance transform for a 3D label array. Use anisotropic parameter to set voxel size.

import edt
import numpy as np

labels = np.array([[[0, 1, 0], [1, 0, 0], [0, 0, 2]]], dtype=np.uint32)
dist = edt.edt(labels, anisotropic=(1.0, 1.0, 1.0))
print(dist)
# Output: 3D distance transform (x, y, z axes)