| Re: Issues with read_png and/or profiler [message #80967] |
Fri, 27 July 2012 13:24 |
hugh.ramp
Messages: 4
Registered: July 2012
|
Junior Member |
|
|
I was reading over a network, although moving it onto local storage did not drastically increase the speed. However, I believe I've found that profiler was the source of misinformation, after editing my code it now shows read_png to take a reasonable amount of time. Instead, I found that the time appropriated to the read_png function was actually from the convol function, which I had been misusing. My code now runs much faster!
Thanks,
Hugh
On Wednesday, 25 July 2012 11:05:14 UTC-6, Brian J. Daniel wrote:
> Are you reading the image over a network connection? If so, your network speed is the bottleneck. Download your file locally before processing.
>
>
>
> Also, check out the convol function. I expect it to take much less time than 1.5 hours that you are reporting.
>
>
>
> -Brian
>
>
>
> On Wednesday, July 25, 2012 11:51:54 AM UTC-4, Hugh wrote:
>
>> Hi all,
>
>>
>
>> I'm running a fairly straight forward image correlation program, which should run fairly quickly (~O(n^4), I believe). However, the complexity seems to be rising much quicker than that, an 962x722 image with a 30x29 kernel takes ~16 hours (with profiler on), whereas a 640x480 image with the same kernel takes ~1.5 hours.
>
>>
>
>> Using profiler to determine the source for the complexity, I found that the Time self(ms) for read_png() was ~60,000,000ms, i.e., 99% of the runtime was loading in the image. However, time+sub(ms) reports taking only ~100ms. I was under the impression that Time+sub should always include time self, no? In any case, I don't think read_png should be taking nearly 16 hours to read a 700,000 pixel image.
>
>>
>
>> I was able to recreate the problem on a separate computer using the same code.
>
>>
>
>> Profiler Snapshot here: http://i.imgur.com/xJelD.png
>
>>
>
>>
>
>> Cheers and thanks,
>
>> Hugh
|
|
|
|
| Re: Issues with read_png and/or profiler [message #80998 is a reply to message #80967] |
Wed, 25 July 2012 10:05  |
Brian Daniel
Messages: 80
Registered: July 2009
|
Member |
|
|
Are you reading the image over a network connection? If so, your network speed is the bottleneck. Download your file locally before processing.
Also, check out the convol function. I expect it to take much less time than 1.5 hours that you are reporting.
-Brian
On Wednesday, July 25, 2012 11:51:54 AM UTC-4, Hugh wrote:
> Hi all,
>
> I'm running a fairly straight forward image correlation program, which should run fairly quickly (~O(n^4), I believe). However, the complexity seems to be rising much quicker than that, an 962x722 image with a 30x29 kernel takes ~16 hours (with profiler on), whereas a 640x480 image with the same kernel takes ~1.5 hours.
>
> Using profiler to determine the source for the complexity, I found that the Time self(ms) for read_png() was ~60,000,000ms, i.e., 99% of the runtime was loading in the image. However, time+sub(ms) reports taking only ~100ms. I was under the impression that Time+sub should always include time self, no? In any case, I don't think read_png should be taking nearly 16 hours to read a 700,000 pixel image.
>
> I was able to recreate the problem on a separate computer using the same code.
>
> Profiler Snapshot here: http://i.imgur.com/xJelD.png
>
>
> Cheers and thanks,
> Hugh
|
|
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
