Chaos Corona Forum
Chaos Corona for 3ds Max => [Max] I need help! => Topic started by: fobus on 2014-12-11, 07:25:14
-
Hi all.
I'm using Corona Alpha v7 and have a question. Is there any trick to get rid of noise in DoF and MB? All is fine with the noise until we catch up any bright spot in our view. It's counless amount of time neede to remove noise from this spots. The only solution I found is use of BUCKET MODE rendering with very agressive settings. But even that giving me very rough bokeh in such areas of images.
Simple example made for this thread.
-
you can try lower the GI/AA balance to 8 or 4 or 2. it help a lot.
and for somecases we have to wait till up to 800-1000passes so that DOF will look smoother.
yes, i think bucket mode is the best for DOF image.. it adaptivity handled noise pretty well.
-
Unfortunately no luck with PROGRESSIVE sampler at all. GI/AA balance set to 2 and 6000 passes with silky smooth GI and bokeh but pronounced noise in areas of interest. Plus huge time for render of course.
-
maybe the problem comes from objects material, can you upload your material settings?
it can be from diffuse albedo, light emiter, caustic enabled on glass material...
-
Scene in first post. Nothing wrong in material in my opinion.
-
You're asking a lot from Corona in this scene :]
If you really need those flying spheres to illuminate your scene maybe it'd be better to replace them with actual Corona lights? See if it helps with noise. Also if it's for animation, i'd rather not turn off image filtering completely, instead i'd choose gaussian filtering and maybe even increase filtering width and/or blurring size.
-
You're asking a lot from Corona in this scene :]
If you really need those flying spheres to illuminate your scene maybe it'd be better to replace them with actual Corona lights? See if it helps with noise. Also if it's for animation, i'd rather not turn off image filtering completely, instead i'd choose gaussian filtering and maybe even increase filtering width and/or blurring size.
This scene is just for reference to this issue. No matter what light sources will be. No matter will it be light sources at all (we can turn light emission off). Bright spots in reflections and lights itself don't want to be blurred (or defocused) smoothly enough.
-
i just downloaded your scene. i'll give it a try tomorrow
-
I manually clean up the specular noise (usually from metals in my case) in DOF in post...I guess the only solution is adaptivity for this ? You can leave it running like that for few days and it will improve just barely.
-
What about animation? You can't leave frame for several days to reduce noise. And no noisebusters can clean it up unfortunately... May be there is another solution?
-
Not sure what others do, but cleaning up stuck up noise in DOF (still mostly specular from highly reflective metals, glasses,etc..) for me can be big pain in Vray as well. I still use DOF in Post for animation.
Btw, you misunderstood me probably :- ) I said you could leave it for days....but there would no little (or no) change, ergo more rendering gets very diminishing returns.
-
three frames of a chrome ball moving FAST (that also reflects very high point ligths) that has motion blur and on top of that DOF?...
I guess rendering is magic so you gonna need some real magic to get away with this.
Imagine 3 chrome balls separated far away reflecting brigth dots. How Corona could do interpolating and blurring between frames?
I would go full cheating.
Maybe there is a way of puting more frames between so there are more to work with (i think Corona calls it "segments" and has a limit according to keymaster RAM issues) for Moblur and doing DOF in post.
Better could be a velocity pass and zpass and cheat the hell of it in post.
NICE EXCERCISE!!!... i'll be playing with scene in the meantime.
-
Post Dof and MB is great for time reduction but has nothing to do when reflections and refractions in play.
-
Lock at my thread.
right now, the only solution is to use bucket mode.
Its adaptivity isn't so bad and the noise is better and uniform than progressive.
But, as we know, it's impossible to know exactly the rendering time/frame like in Progressive
https://forum.corona-renderer.com/index.php?topic=1448.0
-
Even in BUCKET MODE after 30+ minutes there in no clean image.
Is it the end?
-
Lights in your scene are generally too strong, overexposed areas that are strongly motion-blurred are always hard to render. The bucket one seems quite clean to me. Considering amount of motion blur, objects move so fast there's just no way in the hell anyone would notice any noise at that speed, if it's an animation and not a still frame. So either decrease strength of those illuminating materials, increase highlight compression, let it render longer, or leave it the way it is.
I doubt any other renderer would handle it significantly better. If you make scene intentionally very difficult for renderer, and just throw it on it, it will always be hard to render and will take a while. If it's supposed to be a proof that the renderer can't handle something, then that's okay, and well done, but if it's a job you need to finish before deadline, then you've got to think about how to simplify or fake some things, and how to do some compromises, in order to get it done before time runs out.
-
You're right. This scene is too difficult to render in straight way. But this is just an example of things in other scenes with DoF or MB. I have scenes where all is good except this fu...g DoF. And to make pictures clean I need to switch to BUCKET MODE with very agressive setup that leads to huge rendertimes. Plus BUCKET MODE has no DR support, so this huge rendertimes possible to render on one PC only.
-
Yup, as i said, if you make a very difficult scene and just try to throw it on a renderer, it's not gonna work. Or it will, but it will be slow. In such extreme cases, going with GI/AA as low as 1 may be worth it. But i think the best way is to fake it.
-
I hope adaptivity will be in PROGRESSIVE MODE in future. Or, and it will be the best way, we will see Deep compositing possibilities in Corona.
-
It's not bad. This scene is very complex, cause bright lights. I don't know if in VRay you could have better result...
-
I hope adaptivity will be in PROGRESSIVE MODE in future. Or, and it will be the best way, we will see Deep compositing possibilities in Corona.
How's deep compositing related to this?
-
Playing with the scene a lot found the best way to render this.
Edit:
Half way there... a little cheat, but one minute render using Motion Blur and DOF.
Now the issue is isolated to the "strong point lights". Those is better to compose later.
-
One relatively simple way of dealing with stuff like this is to render to double resolution, run denoising there and then scaling down. Typically gives much better results, because you get more leverage for denoising without detail loss. When rendering an animation you can use video denoising plugins like NeatVideo or DE:Noise, which both usually work quite well. You can also denoise via motion vectors but you'll need Fusion or Nuke or something similar to pull that off.
-
yeah neatvideo is great idea for denoise solution. i have spent much time to play with the scene, there were no useful point i get for the result. going with GI/AA as low as 1 like rawalanche said gives me no big different. i think post production is the only one best solution we can reach. but have anyone here try the scene with other renderer?
-
Yes, me. Look at my thread.
borisquezadaa, can you share with us your solution? =)
-
sorry, i didn't open your thread.
and borisquezadaa, what is your little cheat?
-
to_borisquezadaa:
1 minute for frame is very good, but this image has nothing similar with what i'm wanted to reach. This scene is just a reference to global problem. I have many scenes with highlights from different sources and can't reach clean DoF or MB at this areas. Don't think that this particular solution willdo any job at those images. Of course we can find many different solutions in many different scenes, but I started to use Corona for its simplicity and no fake solutions. So this way is not what I expected to use.
to_cecofuli:
I can't see images in Your post unfortunately.
-
Im ashamed... its simple "cheat" but works.
1.- Hide all the chrome balls.
2.- Render lowres (800x400) hemispherical sequence images for all frames.
3.- Create a new material all "zero" except autoilumination, and load the image sequence created in step 2 (make shure that spherical coordinates are used in the sphere) and aply to chromeballs. (Later i think it could work using rayswitcher material. The main idea is to bake the reflections over spheres so corona does not have to calculate those, maybe rayswitch can be a more elegant solution who knows).
4.- Select ligth material -> exclude all chrome spheres.
5.- Hit render.
Also the hotspot ligths can be composed over if you do a "only ligth" pass and instead of save the beauty just save the alpha.
That pass can be composed and you can even change color or exposure in photoshop or after affects... but that is more tricky.
By the way, those ligth streaks over chrome balls can be achieved if instead of saving crappy jpgs as i did you save in a format that has more depth like exrs.
I could not find any examples of what happens with those hot spots under dof and moblur so i dont know for shure if they dim or get stronger.
-
Im ashamed... its simple "cheat" but works.
1.- Hide all the chrome balls.
2.- Render lowres (800x400) hemispherical sequence images for all frames.
3.- Create a new material all "zero" except autoilumination, and load the image sequence created in step 2 (make shure that spherical coordinates are used in the sphere) and aply to chromeballs. (Later i think it could work using rayswitcher material. The main idea is to bake the reflections over spheres so corona does not have to calculate those, maybe rayswitch can be a more elegant solution who knows).
4.- Select ligth material -> exclude all chrome spheres.
5.- Hit render.
no comments
-
Well... at nigth inside a car with polarized glass that pass through at high speed... no one can see the difference.... XD.
The issue is indeed the sampling. (4 minutes beauty, 33 secs light pass compo in photoshop light alpha pass to screen).
-
I think adaptivity in progressive mode could help here. It was available in older versions of Corona and AFAIK it's still accessible somehow via string options but I'm not sure how. You could for example set gi/aa to 2 or 4, let it cook for a while and then adaptivity comes in, it only refines edges with really high noise threshold and it puts A LOT of samples into these pixels.
-
Anyone knows how to enable adaptivity in progressive mode? And why it is hidden?
May be Ondra can help us to find out how to use it...
-
Fobus put the initial samples to something like 6 and try the passes to 6 also, lets see hat you get then, using pretty high initial samples with bucket mode is counter productive I think, think that the increase of passes in bucket mode is not lineal, so you don't have to start with something like 800 passes, that won't help.
Also, can you share a sample scene to try it here?
Cheers!
-
Sample scene in first post.
6 samples and 6 passes give us very rought result. Even 16/8 not acceptable...
-
No. You need LOTS of AA samples for clean DOF. But you don't want those samples for the whole image.
-
Even 6202 passes can't clean that image. I'm trying hard using no cheats at all.
-
I can't clean it with 6202 passes even with GI/AA samples increased to 16 (but it has nothing to do with AA I think). But time to render frame is countless with this setup, so I rendered region only.
-
Only 60.000+ passes can clean it :)
-
Only 60.000+ passes can clean it :)
But, that is still not clean, compared to result i got with mental ray. I wanted to see if indeed any renderer would have a hard time with this scene, but there is no problem with mental ray, at least regarding motion blur and DOF. Mental ray can't reproduce these selfilluminated spheres so there are no these strong reflections of them, which is why i didn't post the result.
EDIT: and those small selfilluminated spheres, even with 60000 passes they look... I don't know what they look like, but this is not what they should look like :D
-
Try with bucket.
-
Try with bucket.
https://forum.corona-renderer.com/index.php/topic,6220.msg42515.html#msg42515
-
I found something useful... if you choose an adequate custom bokeh shape the noise can greatly be reduced it seems the hard edges of default n-sided shapes are too much.
But the pixelated ligths remain. In the reality there would be some kind of corona around.
About 1000 samples (not 6000 so times cut by 6 with custom bokeh). XD. Lunch time... i'll be back.
-
Friends, it 'a lost game. We have to wait for a new Corona version, with optimized adaptivity. That's all ;-)
-
Is this scene available for download somewhere?
-
Post 1... its a great excercise. If you manage to render this... everything else is just a game.
I'm starting to give up. XD.
I know for shure changing anything from main settings only makes it worse.
Bucket mode cleans up a little better but not so much.
Progresive PT/PT not at all.
Fact 1: increasing image filter size to a ridiculous amount blurred the noise but blurries all. (Not good).
Fact 2: Raw power: Distributed render can't deal with animations. So no motion Blur.
Fact 3: External renderfarm: requires at least 2 euros for frame just to make it look usable.(If client pays... ).
Fact 4: Custom bokeh with soft edges improves a little an noise is less evident. (Just dims it).
Fact 5: Only Motion Blur-> no issues. FAST.
Fact 6: Only DOF -> noisy and difficult to clean.
Fact 7: Only motion blur pass with DOF Zdepth-> Photoshop... gross.
Fact 8: Same as 7 in after effects--- > better that photoshop... but gross at the end.
Fact 9: Out of ideas.... anyone?
I think that the myth that Corona can render this in reasonable time is ....
(http://media.tumblr.com/tumblr_li0s9igCBt1qbolbn.jpg)
XD
-
Looks like you lost the DOF somewhere
I guess somewhere in the Custom bokeh
-
borisquezadaa, this is a very hard scene. Not only for Corona, but I think also for VRay, MR etc..
We have to admit (and Ondra also) That Corona isn't optimized to render this scene.
-
Indeed, the DOF makes them impossible. Just checking the SamplerRate on vray side it shows a real nigthmare.
And yes, the custom bokeh was there to "reduce" the dificult dofness... XD.
I totally gave up. This is not possible yet (in reasonable time) and i really try. XD.
Fun to play with do.
-
I had your same conclusion with my old thread.
But my scene wasn't so strong as yours ;-)
https://forum.corona-renderer.com/index.php/topic,1448.0.html
-
Is this scene available for download somewhere?
In first post You'll find initial scene.
-
In first post You'll find initial scene.
I think you need to keep it in the Signature ))
-
In first post You'll find initial scene.
I think you need to keep it in the Signature ))
Done :)
Hope someone find a solution. Or may be even Ondra will have a look at this topic and clear some things.
-
AA, MB and DOF are all sampled once for every camera ray, I would keep GI/AA and lights mult. to minimum allowed as they are sampled per each ray, reduce the `Max Sample Intensity` and maybe use a higher internal resolution with a more blurry filter, but Pavel you do have a hard example to sample without any biased approximations.
From the developers perspective, I think such mixed AA effects could be sampled better, if there was a way to increase the camera ray samples per object, a naive example would be, if the first ray hits a bbox of an object with motion blur, than throw more samples only to that object by a factor of userDefinedValue.