The most wanted feature?

[benjamincillo]

Hi kramon

I'm not expert on which acaptive sampling means

but what I suggest is something to help to clarify how much "clear enough" the image is.

I mean, when rendering large images, and you spend 4 or 5 hours, probably you cannot notice how much difference is, as far image stills computing...probably a 4 hrs image can be considered 95.00% clear from noise, and a 5hrs image is 99% and so on...you never kwno when image is clear enough, it basically depends on what you need

of course, having this option may help to know when an image is ready to use as a test for materials, light and so on.

for samples I usually spend 5-10 min for a 720x405 image, but sometimes noise is very noticeable. If I could know image is at least 75%-80% clear from noise, I would know much much my image samples would take, or a proximately time for rendering a large image, instead being returning back to my pc to check how it is.


I attached a WIP project
it takes only 03:54 min to have an idea about materials and light are, but certainly I dont know how much better it would be in 15 or 20 min rendering

sorry if that sounds kinda crazy... :P

ps, sorry for any english mistakes

[maru]

...so... what is your "wanted feature"?

[lacilaci]

If I understand this correctly you would like to have some clue during rendering to know how long will it take to get to a certain number of passes?

[benjamincillo]

If I understand this correctly you would like to have some clue during rendering to know how long will it take to get to a certain number of passes?

Yes, something that show info about how much time it will take to clear an image from noise, maybe talking about percentage. this way, someone could stop the render when image is 50% -80% clear from noise.

could it be possible Keymaster?

[racoonart]

What you're requesting is simply adaptivity. Technically noise is just the absence of sufficient information - that means a pathtracer would have to collect an infinte number of samples to reach "noise free" images (or in your term "100% clear from noise".
What you want to have is, simplified, a threshold of acceptable value difference (like the dmc noise threshold in vray), which is nothing else than an adaptive approach: "keep sampling the pixel until the noise threshold is reached"

[benjamincillo]

What you're requesting is simply adaptivity. Technically noise is just the absence of sufficient information - that means a pathtracer would have to collect an infinte number of samples to reach "noise free" images (or in your term "100% clear from noise".
What you want to have is, simplified, a threshold of acceptable value difference (like the dmc noise threshold in vray), which is nothing else than an adaptive approach: "keep sampling the pixel until the noise threshold is reached"

Thanks for your simple explanation Deadclown! I Hope this feature could be implemented.

[Juraj]

I sort of think he doesn't necessarily ask for adaptivity, he asks about some direct information on "how" clear/noisy the image, as Laci wrote, a clue on passes :- )

I.E, for example,. how Maxwell groups passes into SL groups, and you have a some estimated values, for example 12 (I don't know, I picked that number randomly, I don't use Maxwell), is enough for interiors,etc.
At the moment, image might be completely clear by 100 passes, at for example 32/2 Gi/Light samples or 200 passes if you use 16/2 settings,etc.. and you basically stop either based on eye evaluation, or by
educated estimate based on previously done renders and test.

He just wants some more exact way to estimate this. But I guess the threshold Deadclown write might be that.

[benjamincillo]

I sort of think he doesn't necessarily ask for adaptivity, he asks about some direct information on "how" clear/noisy the image, as Laci wrote, a clue on passes :- )

He just wants some more exact way to estimate this. But I guess the threshold Deadclown write might be that.

thanks for helping to clarify my request guys

Yes, What I suggest is to have a VISIBLE INFO about how much noise or clear the image is, just in order to know WHEN I could stop the image, depending on my needs

if an image still noisy after 10 min, but info on screen says that is 50% "clear" I can wait 10 minutes more for a better one, just ir order to have a clue about redering times.

It would be use for batch rendering, instead asigning xxx hrs, or XXX passes to all images- which in most cases - are exceeding the needs, user can say: "ok for a 720x405 it may take 5 min for have a 99% "finished" render , and for 5000x3000 it would take  2 hours for a 95% clear image (since waiting for last 5% would be probably an excess, and almost a-not-noticeable quality).

what do you think?

[andreholzmeister]

Of course isosurfaces is not a priority feature, but it is very interesting and it looks like there is no render out there featuring this natively. My votes goes for SSS, hair shader, and better gi with low flickering for animation, since I am a character animation driven artist.

[paxev]

Any plans of adding BVH8 for AVX/AVX2 processors?

[Ondra]

last time I tried it was slower than BVH4. Maybe if the implementation gets/got better over time. But dont expect miracles

[00Ghz]

What about GPU acceleration? :) or even a full GPU version

[zzubnik]

What about GPU acceleration? :) or even a full GPU version

Personally, I think it would be a waste of development time at the moment. GPU rendering has limitations which would need to be worked around.

[00Ghz]

Yes some. but Acceleration and full implementation are 2 things. Any for the latter out of core fixes memory issues, Octane will have it later this year.

Look at Indigo. They get 2-3x speedups by putting some of the workload on the GPU.

[racoonart]

from the main page, in the FAQ (www.corona-renderer.com):

Is Corona CPU or GPU based?
    It is entirely CPU based with no intention of making a GPU version in the foreseeable future. The reason is that the CPU version is fast enough, and there would not be nearly enough speed improvement to justify the amount of changes required and reduced flexibility.