Messiah render speed comparison.

#61

it looks like when Messiah sprays the photons around(like in monte carlo) reflection pass recieves the uninterpolated result. when you filter those shadow hits and interpolate between them to smooth out, they never show up interpolated in the reflection, because the reflection pass has already been made.

i.e. the reflection does not constitute actual geometry that has photon hits, therefore nothing happens to the pixels on the reflection.

if the shadows were baked back onto the scene, then the reflections would show the filtered effect, because the reflection pass is looking at the already filtered shadow.

then again, somewhere in this thread, you all probably said that, so I’m just agreeing with you.
:slight_smile:

fine, I’ll download the test scene and help.

-JoeC

#62

Thanks, found it.

Here’s what I’ve managed to get so far. Can’t upload atm, as my ftp space is down, so it’s a 300x300 image

2min46sec to render on an AthlonXP 3200 ( at 300x300, not 600x600 )

#63

Here’s the project file if anyone wants to look at the settings, although I’ve just noticed that if I render it at 600x600, the colour of the rear wall that’s reflected in the chrome spheres’ changes to a peachy colour, instead of the green / blue colour it was at 300x300. And if you render it at 150x150, the wall / reflection turns blue :hmm:

Anyone have any idea why that’s happening ?

#64

you know that i linked the project file in the first post right?

#65

Hi,

Thank JOECOSMAN !

You resumed all my text in few words !
Next time I will contact you before posting !
lol :)

stooch:

It’s not cheating … it the way a raytracer or scanline(Reyes algorithm) engine works ! You must bake or render (if you don’t like the word “bake”) the shadow before to render the “reflected ray” or final rendering process.
You Must bake, you don’t have to do it, because it’s an hidden process to the user.
If you want to understand how shadows are rendered please read this.
http://www.groovyvis.com/other/raytracing/shadows.html

here is a copy-paste of a sentence:
"
One technique that is sometimes use to accelerate shadow rays is caching… Information about which object is stored."

I did explain how to post process them to make them smoother before the ray-cast computing.

Again, nothing you can handle if you don’t change the behave of the render engine (internal change) to calculate reflected shadows.
No parameters in GUI to fix that.

  • Final gathering ( like in Mentalray is an interpolation process of the “sample rays dots”, it’s not evaluated in the calcul of reflections) to correct that you need to use a lot of photons when it’s not necessary…less phtons could be used if there are no reflections.

Let Fori and Taron code that for us :wink:

#66

im not here to argue with you, the purpose of this thread is render tests…so post em if you got em :slight_smile:

#67

Yes :slight_smile: I just posted mine in case anyone wanted to know what settings I used, that’s all.

#68

Well, baking does have its advantages without a doubt. Especially after I have remembered all the ideas I used to have years and years ago in regards to that. But nowadays I really think it becomes less and less interesting in terms of GI, because the machines are getting so fast that a compromise of this sort is not only not so necessary anymore, but could even expose old weaknesses of renderings. This sort of thing is actually mostly then interesting if nothing in the scene is actually moving or ment to cause a response in the illumination of everything. It’s really a tough one to argue out, because there are many different ways in which baking can be used. THere are additional questions as to what is it going to bake to, which is one of them great and funny questions. Are we generating our own UVmaps for objects that don’t have one, how many objects could this actually handle. Should it be really as simple as making a seperate type of “tool” that can be called to bake a single object like it is or at least used to be in Lightwave for instance?! Hmmm…man…that should really be easy. Hmm…hmmm…hmmm…if that’s all you want, picking an object, specifying which UVmap to use, telling it what format to save the image at and where…yeah, I think that’s really something we could do. I was already thinking a lot further and forgot about that “simple little” thingy. I already envisioned a concept that would allow to automatically generate a whole set of UVmaps for all objects in the scene or selected groups of objects and have them bake at the first rendered frame, saving out all of them to a specified location and that simple rock and roll. That system could suggest another type of photon solution by instead of them using the baked maps…hehehe…ah…the crazy ideas. Just ideas, but fun to share. That’s kind of how it’s like when we work, actually. We get crazy ideas and occasionally even try them out. How fun it would be, if that was all we had to do…hehehe…darned. :rolleyes:

ANyway, we will toss the whole concept of adding a standart baking function into the mix and see what comes out of it. No promisses on how soon it would come, but I’m speaking in favor of it. We’ll see. Wouldn’t it be endlessly better, if it wasn’t necessary at all? Superfast rendering that would making baking nearly slower?!? HAHAHA…now I’m really just teasing, but that’s what I personally would wish for a lot more! :bounce:

#69

for me baking reeks of compromise. you can clearly get the shadows on the floor perfectly smooth without baking in a reasonable amount of time, so whats the point of restricting animation?

how about rendering the shadow on a separate buffer before any reflections are even contemplated. once the shadow is generated and smoothed, then perform the raytracing for the reflections/refractions.

now here is the really useful part.

once you are done rendering…dont discard the shadow buffer! compositors would absolutely love to get just the shadows as a separate pass…

#70

Actually that is exactly what I wanna do. :twisted:

For the panther I did a little trick to speed up fur rendering (cut the rendertimes by 50%), but it requires baked surface shading for the panthers geometry - in animation. I had to do it in Maya. And even there I had to write my own MEL script, because baking animated sequences isn’t supported. :hmm:

Now of course it would have been WAY nicer to just select my panther geometry in Messiah and hit the “BAKE” button. :wink: And even when Messiah comes with fur this baking trick might still be worth it - saving 50% rendertime is a great argument I think.

As I am a comper by heart I can only agree.

#71

but then at the very end. as a last step, it just grained it all up??? WTF???

This problem with “enhanced” was already fixed in one version of messiah (sorry, can’t remember the patch), but reappeared in the latest versions - I was under the impression that some old code slipped back in?

Those advanced baking ideas sound like the lightmaps in Mental Ray 3.4.
Basically they allow you to render maps of a branch of a rendertree into image files with a chosen resolution to disk with a special node and use that maps in that same rendertree from then on. The map can be set to be created only once or overwritten every time (and sure it is a sequence for animations).
This allows for a lot of things that are hard to implement into procedural shaders (like for instance a texture blur) as well as completely user defined on-the-fly baking.
It is also the base of some of the new fast sub surface scattering and translucency shaders.
Very nice concept indeed!

I also wouldn’t forget about the games industry when talking about baking!

Cheers

#72

You know, I was playing with the messiah renderer last night, performing my own tests (basic teapot on a plane with an area light) and saw this strage occurence with the ‘enhanced’ AA mode. I saw the same problem, shadows are being rendered nice and soft, and at the last step, boom, it grained it up, when instead, I was hoping for some further smoothing of shadows.

Even though the ‘enhanced’ mode tends to soften the render, I found it to be kind of pleasant, especially at high resolutions. Kind of gives the soft film-like look, and it also gives a nic(er) AA at low AA Levels and low Shadow Quality values, so it is kind of dissappointing for the AA to improve for the geometry, and for the shadows to de-grade instead.

Of course, the solution is to use the Adaptive SS method instead, but then I have to use higher Level values of SS and Shadow Quality values to achieve the same (or similar) results.

Weird, and I hope this is looked at by pmG.

Anyways, just thought I’d share that with you guys. :slight_smile:

Cheers.

#73

Took 14min 11secs to render ( AthlonXp 3200 ). I had also turned on emmision for the luminous panel in the roof - luminosity@2, emission@1

The reflected radiosity is better on the cubes. But on the spheres reflection, the radiosity on the back wall appears to be back to front - slight red tint beside blue wall edge / blue tint beside red wall edge.

#74

Ya know what, i love the Enhanced mode render look, simply love it, like u said it looks like film, soft… and the AA is infact a lot better in this mode than with Adaptive, where i always get AA artifacts no mater what settings i use.

#75

Very nice! the back wall isnt dark anymore!

#76

Cheers.

Pity about the render time though. There’s no contact shadows under the spheres and the faint GI reflection of the luminous roof light ( onto the back wall in the spheres’ reflections ) is also missing. You can get these back with massive photon numbers ( and changing the roof light back to a panel light ), but then the 'dark back wall / iffy GI reflections on cubes ’ problems come back.

Upped the GI Intensity to 1.75, so this one’s a bit brighter - more like the original Kray render.

#77

I’m sorry but this is pretty close to having no GI and just upping the ambience of the scene. You have to get those darks under the balls in order for it to be a valid solution.

#78

Which can be done, at the cost of the reflections screwing up, just like the tests that yoursefl and stooch have already posted.

I guess stooch could be right, in that there’s possibly a bug in the renderer when photon mapping+GI+reflective surfaces are used together ?

#79

About baking in messiah. Should I say again that baking is really needed for 3d real time ( as well as weight maps ) ?
Have you take a look at microwave : http://www.evasion3d.com/mw_lw_prodinfo.html

This is the kind of features that I am expecting. In the first place you can project ( and theres a lots of options for control ) one object ( High res ) onto another ( low ). of course no need to worry about topology…
Another cool thing is that you can “build” ( with 2 point polys ) your camera projection.
This tool is really great and I am using it all the time.
Again I would like to use messiah but I need some keys features.

Sorry if its off topic.

#80

Guys, it might be worth trying Hybrid instead of Monte Carlo, as it can render the reflected back wall at the correct colour when using a low gather radius, which is needed to get a contact shadow under the spheres.

Just watch the GI Intensity though, as it’ll probably need to be around 0.1 or less. If not, then the image will be washed out. It seems to depend on how many photons you’re using.