Great guide cyberbeing, this is much better.
Also it seems that many are using multi pass filtering with different paramenters, so I wonder if it would be helpful to have a built-in multipass support. For example, tabs titled "run 1""run 2", each has a separate set of paramenters under it, and a preview window capable of showing results of a part or all these steps.
Seems like a lot of code, anyway it's up to petopeto to decide :)
Also it seems that many are using multi pass filtering with different paramenters, so I wonder if it would be helpful to have a built-in multipass support. For example, tabs titled "run 1""run 2", each has a separate set of paramenters under it, and a preview window capable of showing results of a part or all these steps.
Seems like a lot of code, anyway it's up to petopeto to decide :)
the image wasn't affected because (i think) i was using PSCS on a slow computer. now i'm using CS3 on a faster computer and it works. it can magically fix images now.asterixvader said:
well, those who understand this, can tell me why it's not affecting the image i wanna fix?
oh wow, it takes its time to fix the whole image. and it's tiny compared to others.
Updated to work around ATI driver bug. GPU mode should work on ATI cards now.
Updated with a UI update.
The "preview" button is gone; changing values updates the preview immediately.
You can also mousewheel up and down to quickly change settings, and control-mousewheel to change in larger increments.
Overall, this makes experimenting with settings much faster.
To compare two settings, click "compare" on the first settings you want to compare, change values to something else, and right click on the image to flip back and forth between the new and old. (This is a small change to how it worked before--before, you right clicked to show the last settings you previewed.)
The "preview" button is gone; changing values updates the preview immediately.
You can also mousewheel up and down to quickly change settings, and control-mousewheel to change in larger increments.
Overall, this makes experimenting with settings much faster.
To compare two settings, click "compare" on the first settings you want to compare, change values to something else, and right click on the image to flip back and forth between the new and old. (This is a small change to how it worked before--before, you right clicked to show the last settings you previewed.)
Man! YOU A GREAT! THIS PLUG SAVED ME! IT WORKS FINE (PS CS3).
GOOD LUCK!!!
GOOD LUCK!!!
Uh, I still got a question
What's the meaning of different colors in structural tensors view?
What's the meaning of different colors in structural tensors view?
It doesn't even work for me. I must be using the settings wrong.
I don't remember the math, but the colors are a direct translation of per-pixel values that don't translate nicely to colors. It's not really a very useful view.
I don't know what "doesn't work" means, but you should be able to see a difference in any image with noise (screening, JPEG) with the default settings if you zoom in, and it'll get very obvious if -p is turned down.
I don't know what "doesn't work" means, but you should be able to see a difference in any image with noise (screening, JPEG) with the default settings if you zoom in, and it'll get very obvious if -p is turned down.
Structure tensor view is useful for tweaking the amount of detail preserved after filtering. It basically shows you a mask of things which appear to either be excluded completely from filtering or receive reduced filtering.
The red and blue lines are your edges (e.g. hard black outlines on your image). Red lines are horizontal edges. Blue lines are vertical edges.
Green is detail and noise. (you'll usually want to be careful about the amount of green shown, or you will potentially end up with a lot of residue noise, particularly around edges)
If you DO NOT WANT something filtered by Greycstoration, make sure IT IS covered by Red, Blue, or Green in Structure Tensor view.
If you DO WANT something filtered by Greycstoration, make sure IT IS NOT covered by Red, Blue, or Green in Structure Tensor view.
The red and blue lines are your edges (e.g. hard black outlines on your image). Red lines are horizontal edges. Blue lines are vertical edges.
Green is detail and noise. (you'll usually want to be careful about the amount of green shown, or you will potentially end up with a lot of residue noise, particularly around edges)
If you DO NOT WANT something filtered by Greycstoration, make sure IT IS covered by Red, Blue, or Green in Structure Tensor view.
If you DO WANT something filtered by Greycstoration, make sure IT IS NOT covered by Red, Blue, or Green in Structure Tensor view.
I don't think the scaling makes sense, though, so it tends to show up as black or completely saturated, and it's just the first phase of filtering. I find it's easiest to just compare the actual results.
I usually use structural tensors view for tweaking noise scale: just make uniform areas clean and borders green or colored, it's quite quick to get a starting value this way.
I've updated with some memory management changes.
One change causes the preview to scroll around more slowly, but uses less memory on very large images.
I've hit a few rare crashes that I havn't tracked down. I think they're driver-related, but if anyone has trouble with this version, the last release is here.
One change causes the preview to scroll around more slowly, but uses less memory on very large images.
I've hit a few rare crashes that I havn't tracked down. I think they're driver-related, but if anyone has trouble with this version, the last release is here.
I've got a problem here: the newest plugin won't appear in my ps cs3. Reverting to r57 solves this problem. This time I get no entry in eventlog when starting ps cs3.
It's in the "Noise" menu with the other denoise filters.
Got it, thanks. This version can process much larger scans with my system.petopeto said:
It's in the "Noise" menu with the other denoise filters.
Bit of an odd result (scaled down) on the latest version of the plugin on my install of PS CS2. Also had a crash before this with some 6mb crash log. Running greyc again caused the same warped result. I went back to the previous version myself for now, which did do the filtering without the weirdness. Good thing you linked that one too as I accidentally wrote over the file heh oops.petopeto said:
I've updated with some memory management changes.
One change causes the preview to scroll around more slowly, but uses less memory on very large images.
I've hit a few rare crashes that I havn't tracked down. I think they're driver-related, but if anyone has trouble with this version, the last release is here.
When batching in CS4, Greyc will work fine for about 3 images, then throws a bad allocation error. A bug in memory usage, or crappy CS4 architecture? Also, anyway to free ram in between loading images?
Happens to me too, but not when batching, just when I do it one by one.MirrorMagpie said:
When batching in CS4, Greyc will work fine for about 3 images, then throws a bad allocation error. A bug in memory usage, or crappy CS4 architecture? Also, anyway to free ram in between loading images?
In detail:
CS4 starts with 62MB of ram, 0-2% CPU used. I run a pass of greyc(on a 600dpi B5 scan) during which 700MB of ram, 98-100% CPU is used. After it is done, the ram usage goes to 300-320MB and the CPU to 2-6%. Each pass adds on a little more after it finishes, till Photoshop crashes or Greyc throws bad allocation.
Closing the image, Edit > Purge > All, have no effect. Also same thing with CS3
CS4 starts with 62MB of ram, 0-2% CPU used. I run a pass of greyc(on a 600dpi B5 scan) during which 700MB of ram, 98-100% CPU is used. After it is done, the ram usage goes to 300-320MB and the CPU to 2-6%. Each pass adds on a little more after it finishes, till Photoshop crashes or Greyc throws bad allocation.
Closing the image, Edit > Purge > All, have no effect. Also same thing with CS3
Filtering is now run in a separate process. This addresses a number of issues:
- Photoshop's allocator is broken: it's not threadsafe. This required major hacks to work around. Moving processing to a separate process means we can use the standard system allocators, which are completely threadsafe. This is a Photoshop bug; having threadsafe allocators is not a luxury, it's fundamental.
- We're no longer working inside Photoshop's memory space, so most issues related to memory fragmentation should go away. This is probably why some people saw memory use increasing over time.
- It isolates the filter from Photoshop, so if it crashes for some reason, it shouldn't crash Photoshop. Any issues like OpenGL driver bugs shouldn't affect Photoshop much, either.
I also fixed a bug where the preview dialog would sometimes crash if you scrolled outside of the image to the right.
There shouldn't be much visible difference in this version, and no new features are added. This was a fairly large change, so let me know if anything explodes...
- Photoshop's allocator is broken: it's not threadsafe. This required major hacks to work around. Moving processing to a separate process means we can use the standard system allocators, which are completely threadsafe. This is a Photoshop bug; having threadsafe allocators is not a luxury, it's fundamental.
- We're no longer working inside Photoshop's memory space, so most issues related to memory fragmentation should go away. This is probably why some people saw memory use increasing over time.
- It isolates the filter from Photoshop, so if it crashes for some reason, it shouldn't crash Photoshop. Any issues like OpenGL driver bugs shouldn't affect Photoshop much, either.
I also fixed a bug where the preview dialog would sometimes crash if you scrolled outside of the image to the right.
There shouldn't be much visible difference in this version, and no new features are added. This was a fairly large change, so let me know if anything explodes...
It works fine, and solve a problem that the first time of running greyc (use GPU) after starting photoshop will give out weird result
like
but seems the speed of greyc slows down, maybe my delusion..
like
but seems the speed of greyc slows down, maybe my delusion..
I met a problem.
If I'm using new Greyc to handle some big images (like 5000x3500, with 16bit/ch), it won't give out the result, though Greyc-helper.bin keeps calculating something in the background.. it even won't show the progress bar in Photoshop. Nothing wrong with the old version (r57, the lasteat version I can run..).
If I'm using new Greyc to handle some big images (like 5000x3500, with 16bit/ch), it won't give out the result, though Greyc-helper.bin keeps calculating something in the background.. it even won't show the progress bar in Photoshop. Nothing wrong with the old version (r57, the lasteat version I can run..).
Updated with a fix for ATI cards and some error handling fixes.
Tried it out, now works with my ATI HD 5770 correctly at lightning speed. The previous version produced faulty vertical blurs with GPU on.
Yeah, a certain person and I were using the 4870 and 4890 and were getting the same results and told peto about it on irc. (setting aside the fact that I get kernel memory errors when I try to use it on big images)
It's a wondrous thing how peto's able to make his plugin work for hardware he doesn't have. Σ(ﾟДﾟ )
It's a wondrous thing how peto's able to make his plugin work for hardware he doesn't have. Σ(ﾟДﾟ )
Actually, when I turn on gpu for it, the image becomes semi-transparent in CS5.
Locking the transparency beforehand amends the problem though; just a heads up.
Locking the transparency beforehand amends the problem though; just a heads up.
nothing wrong with nVidia (9600gt) + CS5..WtfCakes said:
Actually, when I turn on gpu for it, the image becomes semi-transparent in CS5.
Locking the transparency beforehand amends the problem though; just a heads up.
Yeah i meant it's probably an ati problem; I have a laptop with 9600 and it was working correctly.
Updated with one minor change: "initial gaussian blur" is now only processed on the first pass when multiple iterations are set. Before, "-gauss 1 -iter 2" would run the gaussian blur twice, once for each pass. Now, it only runs once at the start of processing.
midzki
if so, your plugin is broken :3