Function for assigning labels to points #5180

harish-vnkt · 2021-01-11T01:22:00Z

Description

This is a function for the feature request at #5087. It takes an array of coordinates and an image as input and returns a mask where each coordinate is assigned a unique integer label. Thanks in advance for your feedbacks. 😄

Questions/clarifications due to being a first-timer

General design question - The function deals with coordinates on 2 axes only, i.e, fails if coords.shape[1] != 2. Should this be extended to a general case where the coordinates can have arbitrary number of axes as long as the indexing works, i.e, passes as long as coords.shape[1] == image.ndim? @jni
In this function, if the coords are [44, 52], the labelling happens at row=42 and col=52. Is this correct behaviour? I opened a related issue Clarification: coordinate conventions in watershed example #5159 to clarify this for the Watershed Segmentation example

Checklist

Docstrings for all functions
Gallery example in ./doc/examples (new features only)
Benchmark in ./benchmarks, if your changes aren't covered by an
existing benchmark
Unit tests
Clean style in the spirit of PEP8

For reviewers

Check that the PR title is short, concise, and will make sense 1 year
later.
Check that new functions are imported in corresponding __init__.py.
Check that new features, API changes, and deprecations are mentioned in
doc/release/release_dev.rst.

grlee77

Thanks @harish-vnkt, this looks like a great start.

I do think we will want to make this n-dimensional, but fortunately that seems fairly easy to do (see comment).

To avoid ambiguity, we should add a Notes section to the docstring, stating that pixels are considered to fall on integer coordinate locations (starting from zero).

Other considerations are what to do with:
1.) out of range coordinates (we could either explicitly check for this or just relying on NumPy to raise an IndexError. Either way, we should document the behavior add a test case for it)
2.) negative coordinates -> explicitly disallow?
Another alternative to raising errors, could be clipping indices to fall within the image shape.

grlee77 · 2021-01-11T07:24:06Z

skimage/util/points_to_labels.py

+    if coords.shape[-1] != 2 or coords.ndim != 2:
+        raise ValueError("'coords' must be of shape (n, 2)")
+
+    h, w = image.shape[:2]


If we want to support use of this function on multichannel and n-dimensional images, then we should probably add a multichannel keyword argument (defaulting to False). This way we don't have to try and guess based on shape if the image is multichannel. If this multichannel argument is present then, we can get the shape of the spatial dimensions using something like:

shape = image.shape[:-1] if multichannel else image.shape

This can be moved above the shape check above so that you can check if coords.shape[-1] != len(shape) instead of assuming 2 dimensions.

Huh? We shouldn't use multichannel here, all we need is the output shape. ie don't pass in an image as an argument, only the shape.

emmanuelle · 2021-01-11T21:51:56Z

Thank you @harish-vnkt ! In addition to @grlee77 's comments, could you also think of a gallery example which would show how to use this new function?

jni · 2021-01-13T00:40:37Z

skimage/util/points_to_labels.py

+    image: ndarray
+        Image for which the mask is created


The second argument should only be a shape, not an image

jni · 2021-01-13T00:41:17Z

skimage/util/points_to_labels.py

+    coords: ndarray
+        An array of 2D coordinates with shape (n, 2)


Just make this nD:

Suggested change

coords: ndarray

An array of 2D coordinates with shape (n, 2)

coords: ndarray, shape (N, D)

An array of N coordinates of dimension D.

jni · 2021-01-13T00:41:33Z

skimage/util/points_to_labels.py

+    Returns
+    -------
+    label_mask: ndarray
+        A 2D mask of zeroes containing unique integer labels at the `coords`


jni · 2021-01-13T00:42:23Z

skimage/util/points_to_labels.py

+    if not isinstance(coords, np.ndarray):
+        raise TypeError("'coords' must be an ndarray")
+
+    if coords.shape[-1] != 2 or coords.ndim != 2:
+        raise ValueError("'coords' must be of shape (n, 2)")


Remove these checks, instead check only that coords.shape[1] == len(output_shape), and raise a value error saying the dimensionality of the points should match the dimensionality of the image.

jni · 2021-01-13T00:42:39Z

skimage/util/points_to_labels.py

+
+    h, w = image.shape[:2]
+    np_indices = tuple(np.transpose(np.round(coords).astype(np.int)))
+    label_mask = np.zeros((h, w), dtype=np.uint64)


Suggested change

label_mask = np.zeros((h, w), dtype=np.uint64)

label_mask = np.zeros(output_shape, dtype=np.uint64)

jni

@harish-vnkt as @grlee77 said, we do want this to be nD, but it's quite a quick fix! I think between his comments and mine we can get there!

jni · 2021-01-13T00:46:20Z

@emmanuelle instead of a new gallery example, we can use this function in this code cell:

https://scikit-image.org/docs/dev/auto_examples/applications/plot_human_mitosis.html#segment-nuclei

rfezzani

Thank you @harish-vnkt for your contribution.
I think that the file name must be more generic and hidden: skimage/util/_label.py. This keeps door open to adding other util functions in the same vain (who knows 😉).
The label_points function may also be exposed in the skimage.util module (by adding it to skimage/util/__init__.py).

… rename file

pep8speaks · 2021-01-31T18:09:10Z

Hello @harish-vnkt! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-02-01 21:15:01 UTC

harish-vnkt · 2021-01-31T18:12:41Z

Hi guys, sorry for the late updates. I have accounted for all of the changes mentioned in the reviews. I have gone with @jni's suggestion for the output shape argument and updated the code and tests accordingly.

rfezzani

Thank you @harish-vnkt, I left some comments, this is pretty close ;)

skimage/util/_label.py

…to labels, set copy=False when converting `coords` to int

rfezzani

Thank you @harish-vnkt, I left some minor corrections to PEP8, otherwise it looks good for me 🎉

rfezzani · 2021-02-01T16:32:52Z

skimage/util/_label.py

+
+    Notes
+    -----
+    - The labels are assigned to coordinates that are converted to integer and considered to start from 0.


Suggested change

- The labels are assigned to coordinates that are converted to integer and considered to start from 0.

- The labels are assigned to coordinates that are converted to

integer and considered to start from 0.

rfezzani · 2021-02-01T16:34:28Z

skimage/util/_label.py

+    - Negative coordinates raise a ValueError
+    """
+    if coords.shape[1] != len(output_shape):
+        raise ValueError("Dimensionality of points should match the output shape")


Suggested change

raise ValueError("Dimensionality of points should match the output shape")

raise ValueError("Dimensionality of points should match the "

"output shape")

rfezzani · 2021-02-01T16:35:03Z

skimage/util/_label.py

+    if np.any(coords < 0):
+        raise ValueError("Coordinates should be positive and start from 0")
+
+    np_indices = tuple(np.transpose(np.round(coords).astype(np.int, copy=False)))


Suggested change

np_indices = tuple(np.transpose(np.round(coords).astype(np.int, copy=False)))

np_indices = tuple(np.transpose(np.round(coords).astype(np.int,

copy=False)))

jni

Awesome! This looks ready to me. Thank you @harish-vnkt!

contributor made requested changes

harish-vnkt added 3 commits January 10, 2021 01:00

Add function to assign labels to coordinates

0b58346

Include examples to docstring and add error handling

e260bea

Tests for points_to_labels

452d358

harish-vnkt changed the title ~~Points to labels~~ Function for assigning labels to points Jan 11, 2021

grlee77 previously requested changes Jan 11, 2021

View reviewed changes

grlee77 mentioned this pull request Jan 11, 2021

2021's calendar of community management #5169

Closed

harish-vnkt requested a review from grlee77 January 11, 2021 16:26

jni reviewed Jan 13, 2021

View reviewed changes

jni requested changes Jan 13, 2021

View reviewed changes

rfezzani reviewed Jan 21, 2021

View reviewed changes

harish-vnkt added 4 commits January 31, 2021 11:26

Change second argument to output shape and modified code accordingly,…

48fd172

… rename file

Modify tests to reflect argument change, rename file

da83cae

Expose label_points in __init__.py

de7bb1c

Include refactored names in docstring example

bd0f83c

rfezzani reviewed Feb 1, 2021

View reviewed changes

skimage/util/_label.py Outdated Show resolved Hide resolved

skimage/util/_label.py Outdated Show resolved Hide resolved

skimage/util/_label.py Outdated Show resolved Hide resolved

skimage/util/_label.py Outdated Show resolved Hide resolved

Modify check for negative coordinates with np.any, rename label_mask …

b558b2e

…to labels, set copy=False when converting `coords` to int

rfezzani approved these changes Feb 1, 2021

View reviewed changes

Satisfy PEP8

0673bab

harish-vnkt requested a review from jni February 1, 2021 19:55

rfezzani added action: mrg+1 labels Feb 1, 2021

jni approved these changes Feb 1, 2021

View reviewed changes

jni merged commit 7c3c24f into scikit-image:master Feb 1, 2021

jni mentioned this pull request Feb 1, 2021

Feature request: points to labels #5087

Closed

harish-vnkt deleted the points-to-labels branch February 1, 2021 23:13

grlee77 added 🙏 Feature request and removed type: new feature labels Feb 22, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Function for assigning labels to points #5180

Function for assigning labels to points #5180

harish-vnkt commented Jan 11, 2021

grlee77 left a comment

grlee77 Jan 11, 2021

jni Jan 13, 2021

emmanuelle commented Jan 11, 2021

jni Jan 13, 2021

jni Jan 13, 2021

jni Jan 13, 2021

jni Jan 13, 2021

jni Jan 13, 2021

jni left a comment

jni commented Jan 13, 2021

rfezzani left a comment

pep8speaks commented Jan 31, 2021 •

edited

harish-vnkt commented Jan 31, 2021

rfezzani left a comment

rfezzani left a comment

rfezzani Feb 1, 2021

rfezzani Feb 1, 2021

rfezzani Feb 1, 2021

jni left a comment

	label_mask = np.zeros((h, w), dtype=np.uint64)
	label_mask = np.zeros(output_shape, dtype=np.uint64)

	- The labels are assigned to coordinates that are converted to integer and considered to start from 0.
	- The labels are assigned to coordinates that are converted to
	integer and considered to start from 0.

	raise ValueError("Dimensionality of points should match the output shape")
	raise ValueError("Dimensionality of points should match the "
	"output shape")

	np_indices = tuple(np.transpose(np.round(coords).astype(np.int, copy=False)))
	np_indices = tuple(np.transpose(np.round(coords).astype(np.int,
	copy=False)))

Function for assigning labels to points #5180

Function for assigning labels to points #5180

Conversation

harish-vnkt commented Jan 11, 2021

Description

Questions/clarifications due to being a first-timer

Checklist

For reviewers

grlee77 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

emmanuelle commented Jan 11, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jni left a comment

Choose a reason for hiding this comment

jni commented Jan 13, 2021

rfezzani left a comment

Choose a reason for hiding this comment

pep8speaks commented Jan 31, 2021 • edited

Comment last updated at 2021-02-01 21:15:01 UTC

harish-vnkt commented Jan 31, 2021

rfezzani left a comment

Choose a reason for hiding this comment

rfezzani left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jni left a comment

Choose a reason for hiding this comment

pep8speaks commented Jan 31, 2021 •

edited