Mailman 3 April 2018 - scikit-image

rescale with nearest neighbour gives unpredictable results
by Jon Morris April 25, 2018

April 25, 2018

Hello all, Just had a query about transform.rescale (and by extension, warp), which is giving inconsistent results between different platforms. I wondered if this was a bug or was unavoidable. Version info: numpy==1.13.0 scikit-image==0.13.1 scipy==0.19.1 Platforms: Scientific linux 7.4 and Windows 10 So, I have a raster I'm rescaling from 5m resolution to 10m: In [35]: import numpy as np In [36]: input_array=np.array([[ 30.7213501 , 30.73872986, 30.77840255, 30.79360602], ...: [ 30.40055123, 30.39305344, 30.40674613, 30.39675925], ...: [ 30.70790547, 30.67351216, 30.7357262 , 30.70840534], ...: [ 30.3960635 , 30.33706349, 30.38129123, 30.34338268]]) In [37]: from skimage.transform import rescale In [38]: rescale(input_array,0.5,mode='symmetric',preserve_range=True,order=0) Out[38]: array([[ 30.39305344, 30.77840255], [ 30.33706349, 30.70840534]]) When I try the same thing on a different linux server (same distro and version), the output is different. Out[6]: array([[ 30.39305344, 30.40674613], [ 30.33706349, 30.70840534]]) Over on windows, I get a different answer again. Out[125]: array([[ 30.39305344, 30.79360602], [ 30.33706349, 30.34338268]]) It looks like, when downsampling with nearest neighbour interpolation, for each group of four pixels, one is chosen for the output raster, but which one is chosen varies on different platforms. I stepped through the code and the difference happens in the _warp_fast function in _warps_cy, so I couldn't tell exactly why each value was chosen. Is this expected behaviour, or should the same value be chosen on each platform when downsampling? It's messing with our unit tests, as we need a different expected value for each platform we run the tests on. Thanks, Jon Jon Morris Software Developer Skype<sip:jon.morris@jbarisk.com> T +44 (0) 1756 799919 www.jbarisk.com<http://www.jbarisk.com> [Visit our website]<http://www.jbarisk.com> [http://www.jbagroup.co.uk/imgstore/JBA-Email-Sig-Icons-LINKEDIN.png] <https://www.linkedin.com/in/jon-morris-a2897b4/> [Follow us on Twitter] <https://twitter.com/jbarisk>

1 0

WIP / Demo: New maxima finding algorithm(s) for 2D (#3022)
by Lars G. April 22, 2018

April 22, 2018

Dear devs, I'd like to highlight this pull request https://github.com/scikit-image/scikit-image/pull/3022 as a follow-up of an earlier discussion https://mail.python.org/pipermail/scikit-image/2018-April/005514.html on this mailing list. It demonstrates two new algorithms to find local maxima inside an image. Its performance is comparable to `feature.peak_local_max` but has the same output as `morphology.local_maxima`. Please have a look if you're interested. Best regards, Lars

1 0

Re: [scikit-image] arch linux install ?
by K.-Michael Aye April 16, 2018

April 16, 2018

Maybe you don’t need, but I think, these days, it’s highly recommended, also from a point of view of reproducing results betwen Win/Mac/*nix. Happy multi-year conda user here. Best, Michael On Apr 16, 2018, 14:25 +0200, Julien Derr <julien.derr(a)gmail.com>, wrote: > thanks very much Martin and François for your answers, > > I start to understand a little bit more the various options. > But If understand correctlly, it means that under arch linux, If I want to use scikit image, I need a virtual environment ... > > thanks again, > > Julien > > > On Sun, Apr 15, 2018 at 6:47 AM, François Boulogne <fboulogne(a)sciunto.org> wrote: > > > For some weird reasons, you have packages content installed but the > > > package manager didn't mark the package as installed. That's not normal > > > (perhaps because you tried with pip?). > > > > > > Anyway, force the install by passing the option "--force" to yaourt. > > > > > > As a side comment, I usually don't recommend to install libraries with > > > pip on the system as the distribution packages might conflict with them > > > and also for the ease of maintenance. Instead, I prefer to use > > > virtualenvs (for me, with pew). > > > > > > Best, > > > > > > -- > > > François Boulogne. > > > http://www.sciunto.org > > > GPG: 32D5F22F > > > > > > > > _______________________________________________ > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image

1 0

Re: [scikit-image] arch linux install ?
by François Boulogne April 16, 2018

April 16, 2018

Le 16/04/2018 à 14:25, Julien Derr a écrit : > thanks very much Martin and François for your answers, > > I start to understand a little bit more the various options. > But If understand correctlly, it means that under arch linux, If I > want to use scikit image, I need a virtual environment ... > To be more precise, you can install either the package provided by the OS (here, from AUR) or wiht pip a virtualenv, not pip as root. My recommandation works for any linux (or even unix) system, not specifically archlinux. (It's just my 2 cents, then you do as you wish ;) ). Best, -- François Boulogne. http://www.sciunto.org GPG: 32D5F22F

1 0

Re: [scikit-image] arch linux install ?
by François Boulogne April 16, 2018

April 16, 2018

For some weird reasons, you have packages content installed but the package manager didn't mark the package as installed. That's not normal (perhaps because you tried with pip?). Anyway, force the install by passing the option "--force" to yaourt. As a side comment, I usually don't recommend to install libraries with pip on the system as the distribution packages might conflict with them and also for the ease of maintenance. Instead, I prefer to use virtualenvs (for me, with pew). Best, -- François Boulogne. http://www.sciunto.org GPG: 32D5F22F

2 1

region analysis 3d
by imagepy＠sina.com April 15, 2018

April 15, 2018

Hi every one: I want to do some region analysis for 2d/3d block. I saw the skimage.measure.regionprops is very powerful. But if we can add some feature below: 1. I want to got every 3d block's surface area. The regionprops's area for 3d object meas volumn and the perimeter is only2d, I think the method skimage to cound perimeter works for 3d surface, binary_outline it then convolve with a 3*3*3 weight matrix. Another function is march_cubes, it returns a series of face and normal vector, but if we can return a index or a lable array, to indicate which faces are in a same block? 2. I see the regionprops is based on pixels analysis, and in some feature, wo must to copy a new image, I think we can do some work by vector. We can count many feature by contours (except the one need intensity_image). yxdragon

2 1

Re: [scikit-image] arch linux install ?
by François Boulogne April 14, 2018

April 14, 2018

Hi Julien, I'm glad to see you around :) What's the output from the AUR package? I'm not the maintainer of this package but I might be able to help you. I use arch for a while. Best, -- François Boulogne. http://www.sciunto.org GPG: 32D5F22F

2 1

Re: [scikit-image] arch linux install ?
by martin sladecek April 14, 2018

April 14, 2018

Hi Julien, there indeed are other ways of managing python on your system, and the best one is probably conda. https://conda.io/docs/ One of the most useful features of conda is that it separates your system python (packages installed by pacman as dependencies for other programs) and your user python (packages you want to use in your personal projects). Conda can act as a package manager for your user python installation, so you can play around with different versions without messing up your system python. Once you get conda set up, you can install scikit image with: conda install scikit-image Switching to a different way of managing your python installations may seem overwhelming at first, but I promise it will save you a lot of headache in the long run. As an arch user myself, here are a few tips: * Let pacman handle your system python (never use pip as root) * Don't use AUR to install python packages (unless another AUR package depends on them) * Once you start using conda, you might encounter problems when building AUR packages. This is because AUR packages are built in your user environment, but they might require a system python dependency. If you encounter this, just remove conda directory from your PATH and try rebuilding the package. Conda is a well established tool and you'll find plenty of documentation and tutorials for it online. This page should be a good place to start: https://conda.io/docs/user-guide/getting-started.html Best of luck! ~martin On 14/04/18 12:22, Julien Derr wrote: > Hi everyone, > > Is there any arch linux user around ? I was wondering what is the most > straight forward way to install skimage ? > > I tried the AUR repository without success, then I tried PIP with no > success neither > is there a third way, or could you tell me which one you used, (In > this case I will tell you the detail of my errors ...) > > thanks a lot, > > Julien > > > _______________________________________________ > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image

1 0

arch linux install ?
by Julien Derr April 14, 2018

April 14, 2018

Hi everyone, Is there any arch linux user around ? I was wondering what is the most straight forward way to install skimage ? I tried the AUR repository without success, then I tried PIP with no success neither is there a third way, or could you tell me which one you used, (In this case I will tell you the detail of my errors ...) thanks a lot, Julien

1 0

转发：回复：Re: local maxima improvements
by imagepy＠sina.com April 13, 2018

April 13, 2018

Hi，everyone I think we should not use peak_local_max for find watershed's seeds. why not use h_maxima? which can give a h tolerance.I think if we should replace it in the official demo? It would cause a misunderstanding. And scikit-image's h_maxima, h_minima is very slow. here I implements one with numba, https://github.com/Image-Py/imagepy/blob/master/imagepy/ipyalg/hydrology/fi…. you can see if it is useful. yxdragon----- 原始邮件 ----- 发件人：Stefan van der Walt <stefanv(a)berkeley.edu> 收件人："Mailing list for scikit-image (http://scikit-image.org)" <scikit-image(a)python.org> 主题：Re: [scikit-image] local maxima improvements 日期：2018年04月12日 03点08分 On Wed, 11 Apr 2018 12:44:45 +1000, Juan Nunez-Iglesias wrote: > In [7]: image > Out[7]: > array([[ 0., 0., 0., 0., 0., 0.], > [ 0., 1., 0., 0., 0., 0.], > [ 0., 0., 0., 0., 0., 0.], > [ 2., 2., 2., 4., 4., 2.], > [ 2., 2., 2., 4., 4., 2.], > [ 2., 2., 2., 2., 2., 2.]]) > > In [15]: feature.peak_local_max(image) > In [17]: image_peak[tuple(feature.peak_local_max(image).T)] = 1 > > In [18]: image_peak > Out[18]: > array([[ 0., 0., 0., 0., 0., 0.], > [ 0., 1., 0., 0., 0., 0.], > [ 0., 0., 0., 0., 0., 0.], > [ 0., 1., 0., 1., 1., 0.], > [ 0., 1., 0., 1., 1., 0.], > [ 0., 0., 0., 0., 0., 0.]]) That output in column 1 looks highly suspect! This is a great example for a regression test, thanks Yann. Stéfan _______________________________________________ scikit-image mailing list scikit-image(a)python.org https://mail.python.org/mailman/listinfo/scikit-image

3 2