This is a set of assignments with solutions for an upcoming DASH tutorial. It uses a very large astronomical image that is loaded into a block-distributed 2D DASH matrix.
I recommend the 31813x19425 928 MB image 'heic1620a.tif' from http://www.spacetelescope.org/images/heic1620a/. Any other astronomical image in TIFF format will do if it is stored in a 'stripped' manner.
The astro examples need:
- libtiff5-dev
- libsdl1.2-dev (not libsdl2) This is not required but optionally, build with -DWITHSDL in this case.
The assignments are given codes with some DASH data container allocations or operations on them missing. Participants need to fill in few lines of code in order to:
- Declare the 2D block-distributed DASH matrix.
- Copy the images line by line from the libtiff buffer to the distributed matrix using a DASH algorithm.
- First, compute independent local histograms in each local block in parallel.
- Then, sum the local histograms using a DASH algorithm to produce the global result.
- Finally, looking at the brightness histogram define the limit between what is considered as bright pixels and dark pixels.
- Do a quick parallel scan over all pixels to confirm, that the marker color is not found.
- For every pixel, run a given 'checkobject' function. It will report, if it was a bright pixel and, if so, mark all adjacent pixels with the marker color, so that they won't be reported again.
- Count in parallel for every block and combine the result into the global result.
- For simplicity, we don't correct for bright objects at the border between distributed block, which may be counted multiple time. This correction is left as a further exercise ...
The correct number of objects in the 'heic1620a.tif' image is 148498.