[scikit-image] 回复：Re: 回复： Numba on pypi

Thu Jul 13 06:04:17 EDT 2017

Hi ThomasI know local_max has no tolerance, And peak_local_max has some parameter based on distance and abs value,(which the watershed demo in gallery used, but when the area is complex, the result is very massy). the h_max is what I need, but the result is unexpected somewhere, and It is too slow (the same image h_max cost 2.6s, but I wrote in numba cost 0.1s)
May I start a new topic thread? with the test image, and my numba code.
BestYXDragon----- 原始邮件 -----
发件人：Thomas Walter via scikit-image <scikit-image at python.org>
收件人：scikit-image at python.org
抄送人：Thomas Walter <Thomas.Edgar.Walter at googlemail.com>
主题：Re: [scikit-image] 回复： Numba on pypi
日期：2017年07月13日 16点25分

    Hi YXDragon, 

      just a word on some aspect you mention: 

      > the local_max has not a tolerance, so the result is too
      massy, then do a watershed end with too many fragments...,

      The definition that underlies this function is: "A local maximum
      is a region of constant grey level strictly greater than the grey
      levels of all pixels in direct neighborhood of the region." -
      There is no tolerance associated to this definition, and I am
      perfectly fine with this. There are definitely many cases where
      you want to extract all local maxima. Nevertheless, there are
      other functions for detection of maxima that allow one to also
      impose certain criteria (such as h_maxima or peak_local_max).

      A question to Stéfan: would this mean that you would remove all
      cython code from scikit-image or would numba just be another
      option? 

      Best, 

      Thomas. 

      On 7/13/17 8:49 AM, imagepy at sina.com wrote:

      Hi Stéfan:

        I appreciate Numba. for sometimes, we must do a 'for' in
        our python code, but just a 'for' with a 'if', It is fussy to
        compile a so/dll or write cython. Numba is very portable, and
        can run anywhere, just need to install numba and llvmlite. (That
        means our package could be a light-weight library, undepended
        any native so/dll)

        As I metioned befor, many scikit-image's algrisms are
        not exquisite enough(just my own opinion),
            the mid_axi function results too many branch and
        sometimes with hols,  
            the local_max has not a tolerance, so the result is too
        massy, then do a watershed end with too many fragments...,
            and how to build a graph from the skeleton, then do a
        network analysis.

      I want to do a contribute to scikit-image, But after some
        effort, I give up, I prefor to write a dynamic lib rather then
        Cython. In the end, I wrote them in Numba. So I appreciate to
        use Numba.

      Best
      YXDragon

        ----- 原始邮件 -----

          发件人：Stefan van der Walt <stefanv at berkeley.edu>

          收件人：scikit-image at python.org

          主题：[scikit-image] Numba on pypi

          日期：2017年07月13日 14点17分

        Hi everyone,

        As many of you know, speed has been a point of contention in

        scikit-image for a long time. We've made a very deliberate
        decision to

        focus on writing high-level, understandable code (via Python and

        Cython): both to lower the barrier to entry for newcomers, and
        to lessen

        the burden on maintainers. But execution time comparisons, vs
        OpenCV

        e.g., left much to be desired.

        I think we have hit a turning point in the road. Binary wheels
        for

        Numba (actually, llvmlite) were recently uploaded to PyPi,
        making this

        technology available to users on both pip and conda
        installations. The

        importance of this release on pypi should not be dismissed, and
        I am

        grateful to the numba team and Continuum for making that
        decision.

        So, how does that impact scikit-image? Well, imagine we choose
        to

        optimize various procedures via numba (see Juan's blog post for
        exactly

        how impactful this can be:

https://ilovesymposia.com/2017/03/15/prettier-lowlevelcallables-with-numba-jit-and-decorators/).

        The only question we have to answer (from a survival point of
        view)

        needs to be: if, somehow, something happens to numba, will an

        alternative will be available at that time? Looking at the
        Python JIT

        landscape (which is very active), and the current state of numba

        development, I think this is likely. And, if we choose to use
        numba, of

        course we'll help to keep it healthy, as far as we can.

        I'd love to hear your thoughts. I, for one, am excited about the

        prospect of writing kernels as simply as:

        >>> @jit_filter_function

        ... def fmin(values):

        ... result = np.inf

        ... for v in values:

        ... if v < result:

        ... result = v

        ... return result

        >>> ndi.generic_filter(image, fmin, footprint=fp)

        Best regards

        Stéfan

        _______________________________________________

        scikit-image mailing list

        scikit-image at python.org

        https://mail.python.org/mailman/listinfo/scikit-image

      _______________________________________________
scikit-image mailing list
scikit-image at python.org
https://mail.python.org/mailman/listinfo/scikit-image

    -- 
Thomas Walter
27 rue des Acacias
75017 Paris

_______________________________________________
scikit-image mailing list
scikit-image at python.org
https://mail.python.org/mailman/listinfo/scikit-image
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-image/attachments/20170713/5a28e3ef/attachment-0001.html>