DOC: change colormap away from "jet" in examples #1327

endolith · 2014-12-27T19:50:56Z

jet is generally discouraged as perceptually misleading: Replace "jet" as the default colormap matplotlib/matplotlib#875 (comment)
I think it's more clear when all 3 images use the same colormap, to show how an edge detector converts a solid block of brightness into bright lines on just the edges:

          ______
    _____/      \_____  -->  _____/\____/\_____

coveralls · 2014-12-27T20:17:41Z

Coverage remained the same when pulling 47d2ee7 on endolith:patch-1 into f89ab05 on scikit-image:master.

sciunto · 2014-12-28T00:40:01Z

The arguments seem to be convincing. How about the other examples:

doc/examples/applications/plot_coins_segmentation.py:117:ax.imshow(elevation_map, cmap=plt.cm.jet, interpolation='nearest')
doc/examples/applications/plot_rank_filters.py:527:fig.colorbar(ax2.imshow(entropy(image, disk(5)), cmap=plt.cm.jet), ax=ax2)
doc/examples/plot_canny.py:40:ax1.imshow(im, cmap=plt.cm.jet)
doc/examples/plot_edge_filter.py:73:ax3.imshow(edge_scharr - edge_sobel, cmap=plt.cm.jet)
doc/examples/plot_entropy.py:27:img1 = ax1.imshow(entropy(image, disk(5)), cmap=plt.cm.jet)
doc/examples/plot_watershed.py:56:ax1.imshow(-distance, cmap=plt.cm.jet, interpolation='nearest')
viewer_examples/plugins/watershed_demo.py:33:        viewer.ax.imshow(labels, cmap=plt.cm.jet, alpha=0.5)

JDWarner · 2014-12-28T00:56:53Z

👍 I'm on board with never using cm.jet deliberately. It's bad for perception, and unintuitive for examples where people may incorrectly think we used an RGB image rather than mapping color on to a grayscale image.

tacaswell · 2014-12-28T01:16:03Z

As a side note, changing the default color map (and forcing a v2.0) is on the near term horizon for mpl. The sticking point is not if it should be done, but what the new default should be.

JDWarner · 2014-12-28T01:21:41Z

That is fantastic news, @tacaswell!

These still need fixed as they actually specify cm.jet.

tacaswell · 2014-12-28T01:38:59Z

Yeah, I commented before I looked at the changes 🐑 .

endolith · 2014-12-28T16:17:40Z

For the other examples, I'm not sure the best choice

http://scikit-image.org/docs/dev/auto_examples/applications/plot_coins_segmentation.html uses gray for the edge detector, then specifically switches to jet for the elevation map, and then spectral for the markers. Is there a reason for using jet? It looks like spectral was chosen because it conveniently colors a 3-value image with black, gray, and green? (Also this spectral has been renamed to nipy_spectral to distinguish it from the uppercase Spectral.)

Also we could use a different colormap than gray as long as it's one of the nicer ones that increases in brightness with height, like bone or YlGnBu_r.

sciunto · 2014-12-29T01:43:58Z

IMO, gray would do the job instead of jet to show the 8000 effect of a sobel filter.

blink1073 · 2015-01-01T02:58:18Z

I like YlGnBu_r in the short term (for all previous uses of jet), followed by whatever the matplotlib folks settle on for their default. Too bad YlGnBu_r doesn't have a snazzier name. @tacaswell, you may want to consider giving your default map a name that is easy to remember (and say), like jet was.

endolith · 2015-01-02T19:00:38Z

I think viewer_examples/plugins/watershed_demo.py only works with jet? It's using it for labeled colors, not as a heatmap?

There's also \doc\examples\plot_phase_unwrap.py which I don't understand yet

jni · 2015-01-02T21:59:56Z

@endolith I just noticed that! Jet is also inappropriate to show arbitrary label maps. Is there anything it can do? =P

For label maps, I think the best approach is a random colormap in Lab space, like I once did here

endolith · 2015-01-02T22:28:11Z

Well I think that's what the qualitative colormaps are meant for? But they're currently implemented as LinearSegmentedColormap and should have been ListedColormap? matplotlib/matplotlib#881

endolith · 2015-01-02T22:34:15Z

Also I'm confused because it seems like skimage.feature and skimage.filters will be combined into skimage.filter, but the documentation follows this and the actual code doesn't yet? http://stackoverflow.com/a/26877466/125507 Like watershed_demo has from skimage import filters and then edge_image = filter.sobel(viewer.image)?

coveralls · 2015-01-02T23:17:33Z

Coverage remained the same when pulling 3a49af5 on endolith:patch-1 into f89ab05 on scikit-image:master.

jni · 2015-01-02T23:40:02Z

@endolith no, the two modules aren't merging. Canny was moved from filter to feature, and filter will be renamed filters. filter.sobel is a bug, missed in the rename. Can you point me to the exact line in the codebase?

Regarding colormaps, I don't have all that much experience in this space... Is there a qualitative map that can be used for an arbitrary number of segments?

blink1073 · 2015-01-03T02:16:08Z

@endolith, I think we need @tonysyu to weigh in on whether to change to a ListedColormap, since he implemented the LinearColormap.

endolith · 2015-01-03T03:01:54Z

@jni Oh, ok. The one I saw is https://github.com/scikit-image/scikit-image/blob/master/viewer_examples/plugins/watershed_demo.py#L31

grep finds some others to check:

https://github.com/scikit-image/scikit-image/blob/master/skimage/morphology/grey.py#L44
https://github.com/scikit-image/scikit-image/blob/master/skimage/morphology/grey.py#L105
https://github.com/scikit-image/scikit-image/blob/master/skimage/filters/tests/test_filter_import.py#L7

tonysyu · 2015-01-03T03:14:30Z

I think it's a good idea to define some ListedColormaps for this purpose. The LinearColormap is a nice easy way to get color palettes from certain colormaps (e.g. the cm.Set* colormaps), but now that you mention it, a better alternative would be to sample the colormap and save it as a real qualitative colormap; e.g.:

qcmap_jet = ListedColormap(plt.cm.jet(np.linspace(0, 1, 5)))

stefanv · 2015-07-07T23:01:03Z

I guess we need to wait for matplotlib to finalize their decision, but at least we have alternatives now at http://bids.github.io/colormap/

blink1073 · 2015-07-11T09:25:46Z

I think should include the viridia look up table in utils and perhaps set it upon import of skimage if jet is the current setting.

stefanv · 2015-07-11T18:08:37Z

I would not override the matplotlib defaults, in case the user chose to use something different. But we can make sure that, when io.imshow is called and the matplotlib backend is in place, viridis is used. Perhaps viridis can be exposed even more prominently so that users can use it easily?

blink1073 · 2015-07-11T18:11:13Z

skimage.set_cmap?

—
Sent from Mailbox

On Sat, Jul 11, 2015 at 1:08 PM, Stefan van der Walt
notifications@github.com wrote:

I would not override the matplotlib defaults, in case the user chose to use something different. But we can make sure that, when io.imshow is called and the matplotlib backend is in place, viridis is used. Perhaps viridis can be exposed even more prominently so that users can use it easily?

Reply to this email directly or view it on GitHub:
#1327 (comment)

stefanv · 2015-07-11T19:44:13Z

I shy away from carrying state. It often leaves an awful mess in notebooks, e.g.

blink1073 · 2015-07-12T14:50:02Z

How about if we register the cmap on import, and then use it by string literal name by default in imshow and all of our examples?

stefanv · 2015-07-12T15:07:58Z

That works!

endolith · 2015-07-14T00:29:14Z

This PR is just for changing the examples, not the whole package's default.
Yes, some of these should be using ListedColormap instead of LinearSegmentedColormap, but BUG: Convert qualitative colormaps to ListedColormap matplotlib/matplotlib#3973 isn't changing for a while, so there isn't a convenient way to do this in examples yet. Is this patch acceptable as is for now?

blink1073 · 2015-07-14T02:16:59Z

@endolith, this is also about a consistent message throughout the library. We'd like to present a clear alternative to jet that has superior visual perception properties.

blink1073 · 2015-07-17T04:03:43Z

@endolith, it looks like nipy_spectral was added in a later version of matplotlib, which is why the Python 2.7 build is failing.

blink1073 · 2015-07-17T04:05:27Z

Perhaps just use cubehelix instead of nipy_spectral and we'll call it a day on this one?

endolith · 2015-07-17T04:21:00Z

Oh. nipy_spectral is a synonym for spectral that I made to distinguish it from Spectral. In the future it should be used, but if it's making the tests fail I can just change it back.

endolith · 2015-07-17T04:35:41Z

Ok, rebased and removed nipy_spectral for now

stefanv · 2015-07-17T22:02:36Z

Some of these should probably use the new colormap proposed in #1599

soupault · 2016-01-30T17:18:55Z

/ping @endolith
Hi! Could you, please, rebase once again? We should really finalize this.

endolith · 2016-01-31T22:56:27Z

ok, rebased

soupault · 2016-02-01T06:12:11Z

doc/examples/segmentation/plot_watershed.py

@@ -54,7 +54,7 @@

 ax0.imshow(image, cmap=plt.cm.gray, interpolation='nearest')
 ax0.set_title('Overlapping objects')
-ax1.imshow(-distance, cmap=plt.cm.jet, interpolation='nearest')
+ax1.imshow(-distance, cmap=plt.cm.gray, interpolation='nearest')


I'd prefer to use non-grayscale colormap here, as it better presents what is actually the result.

What do you mean?

Nevermind, gray is probably one of the best options here.
I wanted to get terrain-like effect, but to get one the image with higher resolution and histogram equalization are required.

soupault · 2016-02-01T06:20:25Z

Thanks @endolith ! One minor comment from my side and we're ready to go.

soupault · 2016-02-01T06:59:06Z

👍

jni · 2016-02-01T10:08:56Z

Merging this, thank you @endolith!

@scikit-image/core if we want to make some examples use viridis, we can use a different PR for that.

DOC: change colormap away from "jet" in examples

endolith changed the title ~~DOC: change colormap to gray~~ DOC: change colormap away from jet in examples Jul 14, 2015

endolith changed the title ~~DOC: change colormap away from jet in examples~~ DOC: change colormap away from "jet" in examples Jul 14, 2015

endolith force-pushed the patch-1 branch from 3a49af5 to e3bcce1 Compare July 17, 2015 04:32

soupault added 🔧 type: Maintenance Refactoring and maintenance of internals ⏩ type: Enhancement Improve existing features labels Jan 30, 2016

endolith force-pushed the patch-1 branch from e3bcce1 to 7888a85 Compare January 31, 2016 22:45

endolith added 2 commits January 31, 2016 17:53

DOC: Change some jet colormaps to gray

2eb0577

DOC: Change phase example from jet to rainbow

bbc3799

endolith force-pushed the patch-1 branch from 7888a85 to bbc3799 Compare January 31, 2016 22:55

soupault reviewed Feb 1, 2016
View reviewed changes

jni added a commit that referenced this pull request Feb 1, 2016

Merge pull request #1327 from endolith/patch-1

4eee2ee

DOC: change colormap away from "jet" in examples

jni merged commit 4eee2ee into scikit-image:master Feb 1, 2016

endolith deleted the patch-1 branch February 1, 2016 12:39

Uh oh!

DOC: change colormap away from "jet" in examples #1327

DOC: change colormap away from "jet" in examples #1327

Uh oh!

Conversation

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

I would not override the matplotlib defaults, in case the user chose to use something different. But we can make sure that, when io.imshow is called and the matplotlib backend is in place, viridis is used. Perhaps viridis can be exposed even more prominently so that users can use it easily?

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!