jwildfire.org

The official JWildfire forum
It is currently Thu Feb 25, 2021 7:30 pm

All times are UTC+01:00




Post new topic Reply to topic  [22 posts ]  Go to pagePrevious123Next
Author Message
PostPosted:Sat Jul 11, 2015 10:33 pm 
Offline
User avatar

Joined:Thu Mar 20, 2014 2:43 pm
Posts:337
Quote:
Right, it renders internally an image of size x 4.
Can this be user configurable x2, x4, x8? Lets test it.
Quote:
This is not planned, because it is slow
I wish the new one to be at least as slow as the old one! For an intel core2duo is 3 times slower than the old "slow" one, more cores maybe reduce the slowness rate.
Quote:
New one is a cpu killer for me.

Are you using the IR? I think this has then to do with another change (more responsive IR) which can be taken back/made configurable.
It has not to do with IR exclusively,
editor's "Render Preview" is equivalently slow, slow as never before (same with IR), I have to reduce quallity 100 to 30 to get about the same speed,
I dare to say it appears insanely heavy compared to the last one. Someone can use spatial oversampling=1 to get a faster environment but then there is no decent antialias, so I add spatial filter radious=0 for no antialias at all.
Nice to hear that the last one will be abandoned for a (good?) reason, but we have to speed up or config the new one to be more usable and fast


Last edited by pentajazz on Sat Jul 11, 2015 11:01 pm, edited 1 time in total.

Top
   
PostPosted:Sat Jul 11, 2015 10:59 pm 
Offline
User avatar

Joined:Sun May 31, 2015 2:39 pm
Posts:678
With the anti-aliasing I am not sure of what settings to use for different images..... In the flat preview do we want bigger or smaller boxes? Do we want to see more red blocks in the preview or less? Or does it just depend on the image you are working on?
Maybe if you gave some example settings you used with your test images that would help in understanding?


Top
   
PostPosted:Sat Jul 11, 2015 11:24 pm 
Offline
Site Admin
User avatar

Joined:Wed May 15, 2013 10:28 pm
Posts:1508
Quote:
I dare to say it appears insanely heavy compared to the last one.
Please try the new version BETA 2, I have included several improvements


Top
   
PostPosted:Sat Jul 11, 2015 11:30 pm 
Offline
Site Admin
User avatar

Joined:Wed May 15, 2013 10:28 pm
Posts:1508
Quote:
With the anti-aliasing I am not sure of what settings to use for different images..... In the flat preview do we want bigger or smaller boxes? Do we want to see more red blocks in the preview or less? Or does it just depend on the image you are working on?
Maybe if you gave some example settings you used with your test images that would help in understanding?
OK, I will explain this briefly later in the Wiki (and will not answer now).
As a quick idea: the pixels/blocks you see in the model are the influence of raw samples to one pixel of the final image. The more blocks you see, the more raw samples are uses to construct one pixel of your image. Red blocks (or those with low z-coordinate in the 3d-model) mean, that those pixels have a negative impact to the whole, which can cause a sharpening effect.
Having only white blocks usually smoothes things, but does not emphasize details.


Top
   
PostPosted:Sun Jul 12, 2015 1:00 am 
Offline
User avatar

Joined:Thu Mar 20, 2014 2:43 pm
Posts:337
Ok, I have tested beta2, I am writing here just to keep up one topic.
It seems faster, not as fast as before.
I have just realized (as you said) that the new antialias algorithm renders a x2 the actual resolution image (x4 with beta1 ?)!
I am used to render more than 6000x6000 resolution images with 4GB of ram. Now (this version x2?) I cannot even render 4000x4000. "GC overhead limit exceeded" or simply java hangs.
This method eats x2 more ram (or previous beta x4 more ram?) than actually needed before.

For my personal "antialias" methods I prefer to render the highest resolution and then reduce image size as much as I want with some good resizing algorithms or just keep the image for large prints (printer drivers have their way of thinking pixels on paper so antialias sometimes is bad for prints especially on higher than 600dpi prints).
So, the new method is also a RAM killer!

Does this improvement worth all this hardware stress and frustration? Could it be implemented somehow alongside with cleverer ram management? At the time being I will have go back to the previous method.

edit: only "spatial oversampling=1" can use ram as before, but as I said antialias is not worth then, previous is far better.


Top
   
PostPosted:Sun Jul 12, 2015 1:42 am 
Offline
Site Admin
User avatar

Joined:Wed May 15, 2013 10:28 pm
Posts:1508
Quote:
and then reduce image size as much as I want with some good resizing algorithms
This is basically the idea how it works, but it operates on raw samples, not final pixels.
You can turn it off by setting spatial oversampling (which you can think of as internal resolution multiplier) to 1.

Of course, you can turn everything off, render at higher resolution and scale it down. Should even work faster than before.

The only thing which is gone, is the blur (previous "antialiasing"). Just add some blur-variation when needed


Top
   
PostPosted:Sun Jul 12, 2015 2:03 am 
Offline
User avatar

Joined:Thu Mar 20, 2014 2:43 pm
Posts:337
understood. Thanks for the explenations and the blur variation tip. A question, applying to raw samples instead of final pixels does improve image quality or speed or it is just a method to see on going results in IR?


Top
   
PostPosted:Sun Jul 12, 2015 2:45 am 
Offline
Site Admin
User avatar

Joined:Wed May 15, 2013 10:28 pm
Posts:1508
I will explain this later in the Wiki (probably was a fault to do it not together with releasing the stuff, but you know, always not enough time...).
With raw samples I mean the state before mapping it into RGB-space and rounding (currently this is the state you can save as hdr).


Top
   
PostPosted:Sun Jul 12, 2015 12:03 pm 
Offline
User avatar

Joined:Thu Mar 20, 2014 2:43 pm
Posts:337
very well... take as much time you need
After the blur variation tip, I realized the way the antialias used to work in the past, when it could be enabled per transformation, then you turned it to be final only.

reading fractalgee' s similar ram issue I could propose this:

Emulate the old algorithm for no oversampling "blurring antialias"
if oversampling=1 (off) then the user could activate the 2 old params if desirable: "antializing amount" and "antializing radius" and if one wants to deactivate filter kernel use "spatial filter radius = 0" as it used to be. For no oversampling set "spatial filter radius = 0" as default then.

Old renders could also be rendered the same old way again then, someone could need this when rerendering old flames for compatibility
I think this has to be carefully implemented as it appears to be a very hard step forward that could brake things up for many.


Last edited by pentajazz on Sun Jul 12, 2015 6:53 pm, edited 2 times in total.

Top
   
PostPosted:Sun Jul 12, 2015 2:16 pm 
Offline
User avatar

Joined:Thu Mar 20, 2014 2:43 pm
Posts:337
more on this...
I ve tested blur variation to emulate last antializing, it is very close (maybe same) with very small values
I ve used 2 final transformations a preblur+linear and a linear+blur with similar results, tested with a very specific flame problem of mine over pixelated lines that had to be smoother.
I am ok with this as it works...

Now I ve realized that the new antializing is the real thing and has to be there as you have done. But for the compatibility stuff you could generalize a new final blurring variation emulation somehow to be easier, interdependently of the new antializing stuff, as a new ready to use blurring "filter" with params, without ram affection, (disabled by default).
I think of it better over my last proposal. This variation blurring effect is more elegant and clearer sometimes than antialising on pixelated edges. Thanks for sharing this tip.


Last edited by pentajazz on Sun Jul 12, 2015 6:56 pm, edited 1 time in total.

Top
   
Display posts from previous: Sort by 
Post new topic Reply to topic  [22 posts ]  Go to pagePrevious123Next

All times are UTC+01:00


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
cron
Powered by phpBB® Forum Software © phpBB Limited
Theme Created by HOLLYSMOKE
Theme updated to phpBB 3.1.6 by KamijouTouma
Theme Version is 3.1.6-ss2