Playing with this filter during these two days, I found that it is slow. So, to play I had to keep small the test image.
Can the performances be improved?
If the problem lies intrinsically in the process, ignore my request, I don’t ask serious interventions, but if there is some space for an easy improvement, it would be nice (also because the tests imply many trials).
Thanks for the attention.
P.S. forgot to say that it is slow when the smoothness is set to 0 (as I do in my tests currently)
Well, I’ve just tested on my machine, and the speed here is totally acceptable (Ubuntu, 4-cores PC).
Could you share what parameters are slow for you ?
Using the original image (4500x3648 pixels) attached and the paramaters 100,0,4 I had to cancel the process after 30 minutes (half an hour) without the result.
Reducing both width and height to 1/4 (1140x912) I got the result after less than half a minute
Using the same image but with size 1920x1536 it took about 2 minutes to get the result.
Just for a compare with another GMIC filter:
- using cartoon it takes 2 seconds (last input 1920x1536)
Using cartoon on the original image size (4500x3648) it took less than 30 seconds.
IMHO the performance problem depends on the source image being too soft, smooth, without “texture”.
Made an experiment:
- image used the intermediate one (1920x1536)
- applied first Color Abstraction Paint, which creates a great texture
- then applied Crystal: it took a few seconds instead of 2 minutes.
This morning I wanted to double check my “impression” explained in my last opinion.
Downloaded the attached image (original) and run CRYSTAL: result after more than one minute.
“Texturized” the image using whirls and run CRYSTAL: result after 2 seconds.
Hence my two pence tip for people who like to apply CRYSTAL:
- use source images having no large flat spaces or texturize it somehow before running CRYSTAL (obviously the result is different, so much better to use CRYSTAL only on originally well texturized images
).
