Fixed automatic color generation in label2rgb function #4531

mohitpokharna · 2020-03-26T12:19:35Z

Description

Fixed #4507 to now automatically generate colors if number of unique labels exceed the number of default colors. Also added a test function for the same.

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.
Consider backporting the PR with @meeseeksdev backport to v0.14.x

pep8speaks · 2020-03-26T12:19:40Z

Hello @mohitpokharna! 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 2020-03-27 10:23:00 UTC

soupault · 2020-03-26T14:44:54Z

skimage/color/colorlabel.py

+            colors = DEFAULT_COLORS
+            colors = [_rgb_vector(c) for c in colors]
+        else:
+            colors = np.random.random((nlabel, 3))


I see some potential issues with this approach:

It is not reproducible

The contrast between the colors is not guaranteed

The method is prone to color collisions.

I would rather suggest to consider adopting (we don't have it as a hrad dependency anymore) one of the qualitative matplotlib colormaps (https://matplotlib.org/3.2.0/tutorials/colors/colormaps.html) and sample the colors from it.

(Just to say it's not @mohitpokharna fault, @jni suggested it :) #4507 (comment) )

@soupault Thank you for pointing out the potential issues. I have added a fix. Please review.

@mohitpokharna sorry for not being 100% clear. By adopting I meant literally taking a piece of relevant code for the colormap generation from the matplotlib codebase. As scikit-image we have made a decision to switch matplotlib from a requirement to an optional dependency (#2710). Thus, we should avoid importing it in commonly used functions (such as label2rgb). If you have an alternative idea, please, let us know. 👍

@soupault How about indicating colors = [np.asarray(cm.tab20c(x))[:-1] for x in range(n_colors)] in the documentation? The user could make its own color series if needed...

Sorry about all the confusion, everyone! 😅 Actually the np.random.random approach was suggested as an alternative to the user's code, but not something that we would want to do internally. Instead, internally, I think we should use Glasbey + shuffle (+ sampling above some set number of colours, eg 255). But Glasbey is a bit complicated to implement. It might be easier to start from the paper than the code.

@mohitpokharna at any rate, thank you very much for attempting this fix!

@jni Thank you for pointing out. It makes much more sense now. I will try to see if I can carry it on further.

@sciunto not sure if I like that 🙂 . We should rather aim at hiding the complexity.

rfezzani · 2020-04-09T14:32:31Z

skimage/color/colorlabel.py

@@ -1,6 +1,7 @@
 import itertools

 import numpy as np
+from matplotlib import cm


I would be happy to not introduce new dependency to matplotlib

rfezzani

Using a qualitative matplotlib colormap is a good idea and would a great addition for the label2rgb function. But instead of introducing yet an other direct dependency to matplotlib, it would be better to vendor the colormap values (cm.lab20c().colors).
Moreover, I think that interpolating over the colormap values will be necessary to guaranty that all labels' colors are unique.

rfezzani · 2020-04-09T14:55:17Z

skimage/color/colorlabel.py

+    """
+    # tab20c qualitative colormap to generate the colors
+    # Transparency is the last item in the tuple object which is removed
+    colors = [np.asarray(cm.tab20c(x))[:-1] for x in range(n_colors)]


cm.tab20c accepts iterables as input, this means that you can directly use

colors = cm.tab20c(range(n_colors))[:, :-1]

but cm.tab20c is a 20 colors LUT, and colors obtained as you made are not correct:

>>> from pprint import pprint >>> import numpy as np >>> from matplotlib import cm >>> pprint(cm.tab20c(range(25))[:, :-1]) array([[0.19215686, 0.50980392, 0.74117647], [0.41960784, 0.68235294, 0.83921569], [0.61960784, 0.79215686, 0.88235294], [0.77647059, 0.85882353, 0.9372549 ], [0.90196078, 0.33333333, 0.05098039], [0.99215686, 0.55294118, 0.23529412], [0.99215686, 0.68235294, 0.41960784], [0.99215686, 0.81568627, 0.63529412], [0.19215686, 0.63921569, 0.32941176], [0.45490196, 0.76862745, 0.4627451 ], [0.63137255, 0.85098039, 0.60784314], [0.78039216, 0.91372549, 0.75294118], [0.45882353, 0.41960784, 0.69411765], [0.61960784, 0.60392157, 0.78431373], [0.7372549 , 0.74117647, 0.8627451 ], [0.85490196, 0.85490196, 0.92156863], [0.38823529, 0.38823529, 0.38823529], [0.58823529, 0.58823529, 0.58823529], [0.74117647, 0.74117647, 0.74117647], [0.85098039, 0.85098039, 0.85098039], [0.85098039, 0.85098039, 0.85098039], [0.85098039, 0.85098039, 0.85098039], [0.85098039, 0.85098039, 0.85098039], [0.85098039, 0.85098039, 0.85098039], [0.85098039, 0.85098039, 0.85098039]])

as you see, all the colors with index > 20 are equal.
You need to provide values between 0 and 1:

>>> pprint(cm.tab20c(np.linspace(0, 1, 25))[:, :-1]) array([[0.19215686, 0.50980392, 0.74117647], [0.19215686, 0.50980392, 0.74117647], [0.41960784, 0.68235294, 0.83921569], [0.61960784, 0.79215686, 0.88235294], [0.77647059, 0.85882353, 0.9372549 ], [0.90196078, 0.33333333, 0.05098039], [0.99215686, 0.55294118, 0.23529412], [0.99215686, 0.55294118, 0.23529412], [0.99215686, 0.68235294, 0.41960784], [0.99215686, 0.81568627, 0.63529412], [0.19215686, 0.63921569, 0.32941176], [0.45490196, 0.76862745, 0.4627451 ], [0.63137255, 0.85098039, 0.60784314], [0.63137255, 0.85098039, 0.60784314], [0.78039216, 0.91372549, 0.75294118], [0.45882353, 0.41960784, 0.69411765], [0.61960784, 0.60392157, 0.78431373], [0.7372549 , 0.74117647, 0.8627451 ], [0.85490196, 0.85490196, 0.92156863], [0.85490196, 0.85490196, 0.92156863], [0.38823529, 0.38823529, 0.38823529], [0.58823529, 0.58823529, 0.58823529], [0.74117647, 0.74117647, 0.74117647], [0.85098039, 0.85098039, 0.85098039], [0.85098039, 0.85098039, 0.85098039]])

but again, as you see, some colors are equal in this case again. I think that this is not the correct approach...

rfezzani · 2020-04-09T15:14:26Z

skimage/color/tests/test_colorlabel.py

+    rgb = label2rgb(label, image=image)
+
+    # number of unique colors in the rgb image
+    n_colors = len(np.unique(rgb[0], return_counts=True, axis=0)[1])


Suggested change

n_colors = len(np.unique(rgb[0], return_counts=True, axis=0)[1])

n_colors = np.unique(rgb, axis=1).shape[1]

rfezzani · 2020-04-09T15:15:17Z

skimage/color/tests/test_colorlabel.py

+    n_colors = len(np.unique(rgb[0], return_counts=True, axis=0)[1])
+
+    # test if unique colors generated equal the expected number of colors
+    assert_array_equal(n_colors, [exp_colors])


Suggested change

assert_array_equal(n_colors, [exp_colors])

assert n_colors == exp_colors

rfezzani · 2020-04-09T16:47:15Z

Using a qualitative matplotlib colormap is a good idea and would a great addition for the label2rgb function

But @jni suggestion #4531 (comment) is even better ;-)

emmanuelle · 2020-04-26T08:17:25Z

So, how do we proceed here? My feeling is that there are possible long-term improvements consisting in

using the Glasbey LUT
alternatively, using a graph coloring method as suggested in Automatic color generation in the label2rgb function #4507

In the short term, there is the low-hanging fruit of increasing the number of distinct colors between which we loop. We should take am existing discrete/qualitative colormap and vendor it so that we do not increase dependencies. Several possibilities exist with matplotlib's colormaps or plotly's colormaps (I think the Alphabet one is quite nice with its 25 colors). What do you think?

grlee77 · 2021-12-02T19:56:00Z

This relatively new Colorcet package has some nice Glasbey categorical colormaps. It looks like it is fairly lightweight (but does itself have two pure python dependencies: pyct and param)

added functionality for auto generating colors in label2rgb

6bf1d36

pep8 issues fixed

b0b968e

soupault reviewed Mar 26, 2020

View reviewed changes

sciunto requested a review from jni March 27, 2020 06:22

sciunto added the 🩹 type: Bug fix Fixes unexpected or incorrect behavior label Mar 27, 2020

mohitpokharna added 2 commits March 27, 2020 10:00

added colormap to generate colors

e12bdba

Merge branch 'master' into label2rgb-color-gen

8a712a1

sciunto added the 🧐 Needs review label Mar 27, 2020

rfezzani reviewed Apr 9, 2020

View reviewed changes

emmanuelle mentioned this pull request Apr 12, 2020

2020's calendar of community management #4486

Closed

CsatiZoltan mentioned this pull request Dec 7, 2020

Automatic color generation in the label2rgb function #4507

Open

Base automatically changed from master to main February 18, 2021 18:23

mkcor mentioned this pull request Feb 22, 2021

2021's calendar of community management #5169

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fixed automatic color generation in label2rgb function #4531

Fixed automatic color generation in label2rgb function #4531

mohitpokharna commented Mar 26, 2020 •

edited

pep8speaks commented Mar 26, 2020 •

edited

soupault Mar 26, 2020

sciunto Mar 27, 2020

mohitpokharna Mar 27, 2020

soupault Mar 27, 2020

sciunto Mar 27, 2020

jni Mar 30, 2020

jni Mar 30, 2020

mohitpokharna Mar 31, 2020

soupault Apr 4, 2020

rfezzani Apr 9, 2020

rfezzani left a comment

rfezzani Apr 9, 2020

rfezzani Apr 9, 2020

rfezzani Apr 9, 2020

rfezzani commented Apr 9, 2020

emmanuelle commented Apr 26, 2020

grlee77 commented Dec 2, 2021

	n_colors = len(np.unique(rgb[0], return_counts=True, axis=0)[1])
	n_colors = np.unique(rgb, axis=1).shape[1]

	assert_array_equal(n_colors, [exp_colors])
	assert n_colors == exp_colors

Fixed automatic color generation in label2rgb function #4531

Are you sure you want to change the base?

Fixed automatic color generation in label2rgb function #4531

Conversation

mohitpokharna commented Mar 26, 2020 • edited

Description

Checklist

For reviewers

pep8speaks commented Mar 26, 2020 • edited

Comment last updated at 2020-03-27 10:23:00 UTC

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

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

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

rfezzani commented Apr 9, 2020

emmanuelle commented Apr 26, 2020

grlee77 commented Dec 2, 2021

mohitpokharna commented Mar 26, 2020 •

edited

pep8speaks commented Mar 26, 2020 •

edited