Not Compiling on Windows 10 about connected-components-3d

[william-silversmith]

Hi Senthil,

I currently do not have access to a windows machine for compilation unfortunately as my lab runs its software development program on ubuntu and mac. In the past, others have recommended integrating https://www.appveyor.com/ which apparently offers windows builds. Another option is to use mingw as a cross compiler, but I wouldn't be able to verify that those binaries run on my own. It's kind of a pain for me to support windows generally, but this survey https://www.jetbrains.com/research/python-developers-survey-2017/ seems to indicated that based on detected OS of people taking the survey, ~50% of Python users are on windows with ~17% probably on mobile devices and that can be distributed somehow over the other three categories (Windows, Linux, Mac OS). Therefore, I am open to discussing this if it's not overly burdensome. For instance, I can't debug it for you guys on Windows.

Theoretically, you should be able to install on windows anyway using the source distribution if you have a C++ compiler installed and explicitly install numpy first. I'm a bit surprised that your pip install command gave you no matching distribution as there is a source distribution listed here: https://pypi.org/project/connected-components-3d/#files

Would you mind showing me the output of your command?

Thanks!
Will

[william-silversmith]

Hi Senthil,

My friend compiled it for AMD on Python 3.7 for me. Unclear if this will help you, but at least the option is there.

Will

[william-silversmith]

I also found some advice on StackOverflow, you can try using pip install connected-components-3d --no-binary :all: to force install from source.

[tkseneee]

Hi William, Thank you for your reply and tips, but still I can't manage to install it in Windows 10. I will try bit more or else I will migrate to Linux. Actually I segmented all the nodule( small tissue clusters) from the Parenchyma region of Lung CT scan images. Now I need to analyse the nodules which are connected over multiple slices of CT scan. For this I need this 3-d connectivity package. Note: I attached the screenshot of my error with this mail.

…

On Fri, Jan 18, 2019 at 6:20 AM William Silversmith < ***@***.***> wrote: I also found some advice on StackOverflow, you can try using pip install connected-components-3d --no-binary to force install from source. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AowKuh6G2oPuFQYLVYekIrqUrvg2LYUZks5vERpdgaJpZM4aCkDz> .

-- T.K.Senthil Kumar "நாமார்க்குங் குடியல்லோம், நமனை யஞ்சோம்"www.tkseneee.blogspot.com What all we need is not CHANGE, it is AWARENESS.

[william-silversmith]

Hi Senthil,

I think your email attachment got dropped unfortunately. You might have to use the web interface to share photos. Would you mind sharing which Python version you're using and whether you're running on Intel or AMD? Do you have a C++ compiler installed? Maybe I should work on getting AppVeyor working....

If you're in a crunch and there aren't too many labels, you can do the following (it might be kinda slow):

import numpy as np
from scipy.ndimage import measurements

image = create_image()
segids = np.unique(image)[1:] # get rid of 0

output = np.zeros(image.shape , dtype=image.dtype)

num_components = 0
for segid in segids:
    components, num_features = measurements.label( (image == segid) )
    components += num_components
    num_components += num_features
    output += components

You can also adjust the "structure" shape of scipy.ndimage.measurements.label in order to achieve 26-connectivity. By default, I think it is 6-connected.

[tkseneee]

[william-silversmith]

Oh interesting! My colleague just submitted a fix for that last night when he compiled it for windows. Somehow it slipped my mind that you'd have the same issue, so I didn't immediately release a fix. Why don't I do that now.

The issue was that windows needs #include <algorithm.h> in cc3d.hpp.

[william-silversmith]

Ok, I just uploaded some new source files. Give it a try! Version 1.0.1.

[william-silversmith]

There's a little funkiness with the build process so I'm going to release 1.0.2.

• Need to set the __VERSION__ property correctly.
• Numpy 1.16.0 stopped building binaries for Python 3.4

[tkseneee]

Thank you very much. Its working now :)

[william-silversmith]

Fantastic! I hope it's helpful. 😊Good luck with your research!