GNU Astronomy Utilities - Bugs: bug #44888, Convolve on sections of image if...
You are not allowed to post comments on this tracker with your current authentication level.
bug #44888: Convolve on sections of image if input is too large
Submitter: | Mohammad Akhlaghi <makhlaghi> | ||
Submitted: | Tue 21 Apr 2015 02:08:44 AM UTC | ||
Category: | Convolve | Severity: | 3 - Normal |
Item Group: | Crash | Status: | Postponed |
Privacy: | Public | Assigned to: | None |
Open/Closed: | Open |
Wed 27 May 2015 01:24:26 AM UTC, comment #2: |
Mohammad Akhlaghi <makhlaghi>![]() |
Tue 21 Apr 2015 02:08:44 AM UTC, comment #1: This item has been reassigned from the project GNU Astronomy Utilities task tracker to your tracker. |
Mohammad Akhlaghi <makhlaghi>![]() |
Tue 21 Apr 2015 02:08:44 AM UTC, original submission:
When the image size is too large, cut the image into smaller pieces and convolve them then stitch the separate convolved parts to make the output. Since by default Convolve makes heavy use of all the CPU cores in the frequency domain, if the image is too large (in width or height or both), it can take too long and the CPU will heat up to dangerous limits. |
Mohammad Akhlaghi <makhlaghi>![]() |
No files currently attached
Depends on the following items: None found
Items that depend on this one: None found
There are 0 votes so far. Votes easily highlight which items people would like to see resolved in priority, independently of the priority of the item set by tracker managers.
Follow 4 latest changes.
Date | Changed by | Updated Field | Previous Value | => | Replaced by |
---|---|---|---|---|---|
2015-05-27 | makhlaghi | Category | None | ![]() |
Convolve |
Item Group | None | ![]() |
Crash | ||
Status | None | ![]() |
Postponed | ||
2015-04-21 | makhlaghi | Reassign item | GNU Astronomy Utilities, task #13560 | ![]() |
GNU Astronomy Utilities, bug #44888 |
This problem only exists for frequency space convolution.
This can be fixed by using the channel structure, explained in the "Tiling an image" part of the manual. If a channel structure is introduced, then frequency space convolution can be done on each channel independently, thereby decreasing the number of pixels in each row and column, increasing the speed and decreasing the burden on the CPU.
Each row and column that is given to the FFT algorithm can be larger than the channel by half the PSF size. This will allow edge effects to not harm the convolution on the channels that are not lying on the edge of the image.