Eric's GSOC 2009 Webpage
|
BEFORE AND AFTER |
|
Welcome to my GSOC webpage. This has the sole purpose of displaying my results for others to see. The tests will be displayed in five parts:
- small: a large image is downsized using each sampler
- vsmall: a large image is downsized a lot using each sampler
- large: a small image is upsized using each sampler
- vlarge: a small image is upsized a lot using each sampler
- rotate: a small image is rotated by 7 degrees using each sampler
The samplers I have worked on are: upsharp, upsharpfast, upsizefast, upsmoothfast, downsharpfast, downsizefast and downsmoothfast. The image output of existing GEGL samplers is also included for comparison purposes. In order to see the real difference between the samplers, it is best to download the images you would like to see and zoom in using your image viewing software.
SMALL results:
Original Image | |
Upsharp | |
Upsharpfast | |
Upsizefast | |
Upsmoothfast | |
Downsharpfast | |
Downsizefast | |
Downsmoothfast | |
Cubic | |
Lanczos | |
Linear | |
Nearest | |
VSMALL results:
Original Image | |
Upsharp | |
Upsharpfast | |
Upsizefast | |
Upsmoothfast | |
Downsharpfast | |
Downsizefast | |
Downsmoothfast | |
Cubic | |
Lanczos | |
Linear | |
Nearest | |
LARGE results:
Original Image | |
Upsharp | |
Upsharpfast | |
Upsizefast | |
Upsmoothfast | |
Downsharpfast | |
Downsizefast | |
Downsmoothfast | |
Cubic | |
Lanczos | |
Linear | |
Nearest | |
VLARGE results:
Original Image | |
Upsharp | |
Upsharpfast | |
Upsizefast | |
Upsmoothfast | |
Downsharpfast | |
Downsizefast | |
Downsmoothfast | |
Cubic | |
Lanczos | |
Linear | |
Nearest | |
ROTATE results:
Original Image | |
Upsharp | |
Upsharpfast | |
Upsizefast | |
Upsmoothfast | |
Downsharpfast | |
Downsizefast | |
Downsmoothfast | |
Cubic | |
Lanczos | |
Linear | |
Nearest | |
Other GEGL stuff that I worked on:
- I worked on a hack solution to removing the transparent border that appears around images produced by samplers. By changing the sampler initialization values to (0,0,1,1) the border is greatly reduced. The sampler's actual values are set at the beginning of the "get" call.
- I looked into a solution that would allow users to set their own abyss color via their XML test files. It was determined that the solution would be too time consuming to implement and too much internal code would have to be modified for this to be possible.
- I created a patch that removes the YAFR sampler. It was determined that this method is inferior to newer sampler implementations.
- I investigated some compile warnings with dependencies of the affine.c file. This turned out to be an issue with the generation of documentation at compile time.
Links to bugzilla bugs that I contributed to: