Vladimir, there are still problems with the AA - look at the upside down chair legs and window frames. Did it really take you only 30 min of tweaking each lights/diffuse/glossy samples test, then AA samples test, if the final render took 44 min? Where you rendering the tests at smaller rez?
@Tim, there is no type of AA sampling selector in Maxwell - except a slider for "softer" vs "sharper" looking AA. Slider goes from 0 to 100 and I used 60 (the default). This can be changed during or after the render. The difference is subtle, so below I've saved 3 different versions - default 60, 0 and 98. We use a proprietary AA sampling technique which was a big part of V2 of Maxwell :) @Emilio - there isn't really that much you can do to optimize a scene like this in Maxwell or any unbiased renderer that doesn't have a built in hard bounce limit. This type of scene is one of the hardest things you can throw at them. One of the things you can do if you can however is not mix normal emitters with sun or skydome since these two types of emitters have very different strengths and Maxwells optimisations don't work well in these cases. Here is the classroom scene with Maxwell materials applied (ceiling emitters are hidden): https://dl.dropbox.com/u/32548637/Classroom_test.rar Here is another example with Physical sky this time, with the ceiling emitters hidden. I've also made the floor glossy reflective instead of just diffuse which should add to the complexity, but the render time was only 3h for this one to reach the same sampling level. There are some small more noisier areas on the floor - those are in fact glossy caustics from light reflecting from the low glossiness chrome legs onto the glossy floor. It's nice to know you have this subtlety if you want it, but you can turn indirect or also direct caustics off in the render settings to speed things up. AA 0: https://www.dropbox.com/s/653hesfhbwjw0ln/class3h_Sun_AA0.png AA default: https://www.dropbox.com/s/ki5ji9nlp5ecvn2/class3h_Sun_AAdefault.png AA 98: https://www.dropbox.com/s/jxr0f2zn9jyo6k9/class3h_Sun_AA98.png On Mon, Feb 18, 2013 at 12:06 AM, Vladimir Jankijevic < vladi...@elefantstudios.ch> wrote: > it's so clear to me that I'm using arnold that I sometimes forget to > mention it. sorry about that :/ > settings depend very much on how you approach such a render. here is a > short overview on how I did it: > the one with DoF needs more AA samples so the way I do it is to raise the > AA to 10. Then I dial every aspect (lights, diffuse, glossy) of the > sampling separate until I have the same amount of noise in each of them. I > start with one sample and raise it until I'm somehow satisfied. Mostly 2 > samples per component is enough and if we have heavy contrast in the glossy > component I raise that to 3. Then raise the AA slowly to the expected noise > level. In this case to 20 to really smooth things out. So here are the > sampling settings: > AA: 20 > Diffuse: 2 > Glossy: 3 > one QuadLight in front of the windows: 2 > Sharper Distant Light: 1 > Raydepth for all components to 1 but diffuse to 4 > > the one without DoF doesn't need such high AA samples so we > can rearrange the sample settings to something like this: > AA: 6 > Diffuse: 7 > Glossy: 6 > one QuadLight in front of the windows: 5 > Sharper Distant Light: 4 > Raydepth for all components to 1 but diffuse to 4 > > > here is another one without glossy rays. this one brings the render down > to 27min. it's the same setup as the one with DoF except without any glossy > rays shot: http://xsi.jankin.com/dump/ClassRoom_a.0001_27min.jpg > > > Cheers > Vladimir > > > On Sun, Feb 17, 2013 at 2:26 PM, Steven Caron <car...@gmail.com> wrote: > >> @vladimir, arnold correct? >> >> indeed its good, some of the other renders dont have the proper >> light/shadow on the ground. i think its because of the techniques being >> used on the windows. the 'lightplane' mesh, if its being used as an area >> source it tends to give improper shadowing on the floor. >> >> so talking about technique, whats your setup like and sample settings? >> all my renders still come in around 2hrs with too much noise still. >> >> s >> >> >> On Sun, Feb 17, 2013 at 8:16 AM, Vladimir Jankijevic < >> vladi...@elefantstudios.ch> wrote: >> >>> alright, here is my try on the scene. I tried to get it to a level of >>> the 7h maxwell render. >>> the maxwell render was done on a i7 2600K @ 3.4GHz and mine was on a X5680 >>> @ 3.33GHz. the difference factor is about 2.75. So on my machine the >>> maxwell one would have taken 2h32min >>> so the first render was with depth of field and took 44min at 960x540px >>> the second without DoF took 30min at the same resolution. >>> I'm quiet happy with the outcome for such a good quality render and low >>> setup time (about 30min from scratch), especially knowing that I can now >>> throw what ever I want in that scene and in whatever quantity and it will >>> render :) >>> >>> with DoF: http://xsi.jankin.com/dump/ClassRoom_a.0001_44min.jpg >>> without: http://xsi.jankin.com/dump/ClassRoom_a.0001_30min.jpg >>> >>> >>> cheers >>> Vladimir >>> >>> >>> On Sat, Feb 16, 2013 at 5:58 PM, Steven Caron <car...@gmail.com> wrote: >>> >>>> the camera in the arnold version of the scene file ronald provided >>>> should match... >>>> >>>> https://dl.dropbox.com/u/4271217/Arnold_Classroom.rar >>>> >>>> >>>> On Sat, Feb 16, 2013 at 2:54 PM, Octavian Ureche <okt...@gmail.com>wrote: >>>> >>>>> Here's another one. Tried to match the angle and res of the other >>>>> renders (as best as i could). >>>>> Kept the green "lcd" and the burned-out look simply out of preference. >>>>> Had to up the sub-pixel sampling to counter the artifacts in the sharp >>>>> reflective metals (could have pushed it further). >>>>> But it balanced itself nicely with the lowered res, and got >>>>> approximatively the same rendertime. >>>>> Manged to get it to render in under 7 mins with lower glossy samples, >>>>> but it was a bit noisier for my taste. >>>>> >>>>> Buenas noches, >>>>> Octav >>>>> >>>> >>>> >>> >>> >>> -- >>> --------------------------------------- >>> Vladimir Jankijevic >>> Technical Direction >>> >>> Elefant Studios AG >>> Lessingstrasse 15 >>> CH-8002 Zürich >>> >>> +41 44 500 48 20 >>> >>> www.elefantstudios.ch >>> --------------------------------------- >>> >> >> > > > -- > --------------------------------------- > Vladimir Jankijevic > Technical Direction > > Elefant Studios AG > Lessingstrasse 15 > CH-8002 Zürich > > +41 44 500 48 20 > > www.elefantstudios.ch > --------------------------------------- >
