| Re: Segfault when smoothing image [message #69771] |
Mon, 15 February 2010 07:08  |
thoeger
Messages: 4
Registered: February 2010
|
Junior Member |
|
|
On Feb 15, 1:36 pm, Jeremy Bailin <astroco...@gmail.com> wrote:
> On Feb 15, 6:05 am, thoeger <lusepus...@gmail.com> wrote:
>
>
>
>
>
>> Sorry if I'm not precise enough;
>
>> The data I have is an array that is later to be exported to an image.
>> The array is generated by simply making a 2200x2200 float array of
>> 0's, and then changing the values to 1.0 at certain coordinates that
>> are given in a list of objects in a different image. That gives a 2D
>> array of 0's with a few 1's here and there, which is then to be
>> smoothed, so each pixel gets a value between 0 and 1, depending on how
>> far they are from the pixels originally of value 1.
>
>> The code line I have used is:
>
>> imgdata2 = filter_image( imgdata, FWHM_GAUSSIAN=300, /
>> ALL_PIXELS)
>
>> imgdata being the name of the array I've previously generated.
>> From GDL, I receive the messages
>
>> % Compiled module: FILTER_IMAGE.
>> % Compiled module: FACTOR.
>> % Compiled module: PRIME.
>> % Compiled module: PSF_GAUSSIAN.
>> % Compiled module: GAUSSIAN.
>> % Program caused arithmetic error: Floating underflow
>> % Compiled module: CONVOLVE.
>> Segmentation fault
>
>> and then it exits. In real IDL, it simply stalls indefinitely after "%
>> Compiled module: GAUSSIAN."
>
>> On Feb 14, 6:51 pm, Gianguido Cianci <gianguido.cia...@gmail.com>
>> wrote:
>
>>> What is your input for all this, a bunch of x,y coords? And you want a
>>> certain value at each coord in a "fake" 2200x2200 iamge?
>>> Could you post examples of input and especially the code you are
>>> using?
>
>>> --Gianguido
>
>>> On Feb 14, 4:54 am, thoeger <lusepus...@gmail.com> wrote:
>
>>>> Hello newsgfroup;
>
>>>> I hope this question isn't too basic.
>
>>>> As part of my master thesis in astronomy, I have to make an image
>>>> consisting of 2200x2200 pixels having the value zero except certain
>>>> pixels, representing each the center of tan astronomical objects,
>>>> having the value one. The goal is to get an idea of the number density
>>>> of objects in the field, so I try to do a gaussian smoothing using
>>>> the filter_image function, but end up with a segfault and IDL
>>>> quitting due to floating underflow. (To be precise, this is GDL on my
>>>> laptop. True IDL on the university computers just stalls
>>>> indefinitely). So it seems I'm doing something wrong here. Does anyone
>>>> have an idea how to implement the smoothing, if not by filter_image?
>
> I'm not sure what FILTER_IMAGE is doing internally, but depending on
> what method it uses it could be very memory-intensive. In
> POINT_CONVOLVE, you can pick from a few different methods, some of
> which will be more efficient in different circumstances.
>
> Unfortunately, I forgot to renew the domain name... I'll post a link
> to POINT_CONVOLVE when it's back up and running. ;-) But if you
> already have JBIU, it's in there.
>
> -Jeremy.
Jeremy, that looks very interesting and promising. Waiting in
anticipation. :-)
|
|
|
|
| Re: Segfault when smoothing image [message #69772 is a reply to message #69771] |
Mon, 15 February 2010 04:36  |
Jeremy Bailin
Messages: 618
Registered: April 2008
|
Senior Member |
|
|
On Feb 15, 6:05 am, thoeger <lusepus...@gmail.com> wrote:
> Sorry if I'm not precise enough;
>
> The data I have is an array that is later to be exported to an image.
> The array is generated by simply making a 2200x2200 float array of
> 0's, and then changing the values to 1.0 at certain coordinates that
> are given in a list of objects in a different image. That gives a 2D
> array of 0's with a few 1's here and there, which is then to be
> smoothed, so each pixel gets a value between 0 and 1, depending on how
> far they are from the pixels originally of value 1.
>
> The code line I have used is:
>
> imgdata2 = filter_image( imgdata, FWHM_GAUSSIAN=300, /
> ALL_PIXELS)
>
> imgdata being the name of the array I've previously generated.
> From GDL, I receive the messages
>
> % Compiled module: FILTER_IMAGE.
> % Compiled module: FACTOR.
> % Compiled module: PRIME.
> % Compiled module: PSF_GAUSSIAN.
> % Compiled module: GAUSSIAN.
> % Program caused arithmetic error: Floating underflow
> % Compiled module: CONVOLVE.
> Segmentation fault
>
> and then it exits. In real IDL, it simply stalls indefinitely after "%
> Compiled module: GAUSSIAN."
>
> On Feb 14, 6:51 pm, Gianguido Cianci <gianguido.cia...@gmail.com>
> wrote:
>
>
>
>> What is your input for all this, a bunch of x,y coords? And you want a
>> certain value at each coord in a "fake" 2200x2200 iamge?
>> Could you post examples of input and especially the code you are
>> using?
>
>> --Gianguido
>
>> On Feb 14, 4:54 am, thoeger <lusepus...@gmail.com> wrote:
>
>>> Hello newsgfroup;
>
>>> I hope this question isn't too basic.
>
>>> As part of my master thesis in astronomy, I have to make an image
>>> consisting of 2200x2200 pixels having the value zero except certain
>>> pixels, representing each the center of tan astronomical objects,
>>> having the value one. The goal is to get an idea of the number density
>>> of objects in the field, so I try to do a gaussian smoothing using
>>> the filter_image function, but end up with a segfault and IDL
>>> quitting due to floating underflow. (To be precise, this is GDL on my
>>> laptop. True IDL on the university computers just stalls
>>> indefinitely). So it seems I'm doing something wrong here. Does anyone
>>> have an idea how to implement the smoothing, if not by filter_image?
I'm not sure what FILTER_IMAGE is doing internally, but depending on
what method it uses it could be very memory-intensive. In
POINT_CONVOLVE, you can pick from a few different methods, some of
which will be more efficient in different circumstances.
Unfortunately, I forgot to renew the domain name... I'll post a link
to POINT_CONVOLVE when it's back up and running. ;-) But if you
already have JBIU, it's in there.
-Jeremy.
|
|
|
|
| Re: Segfault when smoothing image [message #69773 is a reply to message #69772] |
Mon, 15 February 2010 03:05 |
thoeger
Messages: 4
Registered: February 2010
|
Junior Member |
|
|
Sorry if I'm not precise enough;
The data I have is an array that is later to be exported to an image.
The array is generated by simply making a 2200x2200 float array of
0's, and then changing the values to 1.0 at certain coordinates that
are given in a list of objects in a different image. That gives a 2D
array of 0's with a few 1's here and there, which is then to be
smoothed, so each pixel gets a value between 0 and 1, depending on how
far they are from the pixels originally of value 1.
The code line I have used is:
imgdata2 = filter_image( imgdata, FWHM_GAUSSIAN=300, /
ALL_PIXELS)
imgdata being the name of the array I've previously generated.
From GDL, I receive the messages
% Compiled module: FILTER_IMAGE.
% Compiled module: FACTOR.
% Compiled module: PRIME.
% Compiled module: PSF_GAUSSIAN.
% Compiled module: GAUSSIAN.
% Program caused arithmetic error: Floating underflow
% Compiled module: CONVOLVE.
Segmentation fault
and then it exits. In real IDL, it simply stalls indefinitely after "%
Compiled module: GAUSSIAN."
On Feb 14, 6:51 pm, Gianguido Cianci <gianguido.cia...@gmail.com>
wrote:
> What is your input for all this, a bunch of x,y coords? And you want a
> certain value at each coord in a "fake" 2200x2200 iamge?
> Could you post examples of input and especially the code you are
> using?
>
> --Gianguido
>
> On Feb 14, 4:54 am, thoeger <lusepus...@gmail.com> wrote:
>
>
>
>> Hello newsgfroup;
>
>> I hope this question isn't too basic.
>
>> As part of my master thesis in astronomy, I have to make an image
>> consisting of 2200x2200 pixels having the value zero except certain
>> pixels, representing each the center of tan astronomical objects,
>> having the value one. The goal is to get an idea of the number density
>> of objects in the field, so I try to do a gaussian smoothing using
>> the filter_image function, but end up with a segfault and IDL
>> quitting due to floating underflow. (To be precise, this is GDL on my
>> laptop. True IDL on the university computers just stalls
>> indefinitely). So it seems I'm doing something wrong here. Does anyone
>> have an idea how to implement the smoothing, if not by filter_image?
|
|
|
|
| Re: Segfault when smoothing image [message #69776 is a reply to message #69773] |
Sun, 14 February 2010 09:51 |
cgguido
Messages: 195
Registered: August 2005
|
Senior Member |
|
|
What is your input for all this, a bunch of x,y coords? And you want a
certain value at each coord in a "fake" 2200x2200 iamge?
Could you post examples of input and especially the code you are
using?
--Gianguido
On Feb 14, 4:54 am, thoeger <lusepus...@gmail.com> wrote:
> Hello newsgfroup;
>
> I hope this question isn't too basic.
>
> As part of my master thesis in astronomy, I have to make an image
> consisting of 2200x2200 pixels having the value zero except certain
> pixels, representing each the center of tan astronomical objects,
> having the value one. The goal is to get an idea of the number density
> of objects in the field, so I try to do a gaussian smoothing using
> the filter_image function, but end up with a segfault and IDL
> quitting due to floating underflow. (To be precise, this is GDL on my
> laptop. True IDL on the university computers just stalls
> indefinitely). So it seems I'm doing something wrong here. Does anyone
> have an idea how to implement the smoothing, if not by filter_image?
|
|
|
|
| Re: Segfault when smoothing image [message #69861 is a reply to message #69773] |
Mon, 15 February 2010 09:45 |
wlandsman
Messages: 743
Registered: June 2000
|
Senior Member |
|
|
On Feb 15, 6:05 am, thoeger <lusepus...@gmail.com> wrote:
> The code line I have used is:
>
> imgdata2 = filter_image( imgdata, FWHM_GAUSSIAN=300, /
> ALL_PIXELS)
> In real IDL, it simply stalls indefinitely after "%
> Compiled module: GAUSSIAN."
(filter_image.pro is from the IDL Astronomy Library
http://idlastro.gsfc.nasa.gov/ftp/pro/image/filter_image.pro )
Let's look at the number of arithmetic operations you are asking
for. By default, FILTER_IMAGE uses a kernel 3 times the size of the
FWHM, which in your case would be a 900 x 900 array. The /
ALL_PIXELS option reflects pads the main array to avoid edge effects,
so your initial 2200 x 2200 array will become a 3100 x 3100 array.
The number of multiplications for the convolution is then
900.^2 * 3100.^2 = 7.8e12
As Jeremy mentioned, one probably wouldn't use a direct convolution
for a problem like this. By default, FILTER_IMAGE computes the
convolution as a product of Fourier transforms, but it appears that
this is still too slow. A faster method is to approximate the
Gaussian convolution using iterated 3 x 3 SMOOTHing.
b = filter_image(a,smooth=300,/iterate)
(This will automatically take care of edge effects.)
There is a nice tutorial on the different smoothing at
http://www.jhlabs.com/ip/blurring.html (though it is in Java and they
call it blurring).
--Wayne
> On Feb 14, 6:51 pm, Gianguido Cianci <gianguido.cia...@gmail.com>
> wrote:
>
>> What is your input for all this, a bunch of x,y coords? And you want a
>> certain value at each coord in a "fake" 2200x2200 iamge?
>> Could you post examples of input and especially the code you are
>> using?
>
>> --Gianguido
>
>> On Feb 14, 4:54 am, thoeger <lusepus...@gmail.com> wrote:
>
>>> Hello newsgfroup;
>
>>> I hope this question isn't too basic.
>
>>> As part of my master thesis in astronomy, I have to make an image
>>> consisting of 2200x2200 pixels having the value zero except certain
>>> pixels, representing each the center of tan astronomical objects,
>>> having the value one. The goal is to get an idea of the number density
>>> of objects in the field, so I try to do a gaussian smoothing using
>>> the filter_image function, but end up with a segfault and IDL
>>> quitting due to floating underflow. (To be precise, this is GDL on my
>>> laptop. True IDL on the university computers just stalls
>>> indefinitely). So it seems I'm doing something wrong here. Does anyone
>>> have an idea how to implement the smoothing, if not by filter_image?
>
>
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
