Forest Aerial

Started by miqtidar, November 18, 2012, 08:43:35 AM

Previous topic - Next topic

miqtidar

Forest Aerial

Hey guys here is my new Forest Aerial TG Animation.


http://vimeo.com/53754092


It took 1273 Hours to render on my 2 i7 3770k computers.
Kindly see the Image captures below for details.

Hardware Used
Intel i7 3770k 12gb Ram with western Digital Hard Drive.
2 Computers with Clone hardware configs.


Vegitation used in the Project
Xfrog Library,Tree (ML01 Cedrus atlantica Atlas Cedar)
And Mr Lamppost's Basic Bush Pack
Mr Lampposts Basic Grass Pack

Kindly turn HD On for good quality.


Here is the setting of the file.

Number of frame 288
Atmosphere sample = 82
Render Camera Motion blur length = 0.2
Max bucket size = 128 128
Quality Detail = 0.75
AA = 16 (Ray trace objects checked)
Pixel Filter= Mitchell-Netravali
Detail Blending = 1
Displacement filter =1
Microvertex jittering (checked)
Ray Detail Region = 2
Detail In Camera selected
GI Prepass Padding = 0
Do ray trace shadows (checked)
Gi Relative Detail =2
Gi Relative Quality = 2
Gi Blur Radius =8
Read Gi cache to (Interpolate for animation) Number of files to blend 2.
36 Gi cache files were rendered for whole sequence.


Thanks for watching,

Miqtidar

Simius Strabus

Good looking animation you got there. Like it :)
Why stop dreaming when you wake up?

iMac i7 2.8GHz 8Gb

miqtidar


Henry Blewer

I liked this also. I would have added a bit more camera panning, though I admit, it is hard to 'see' the videos end result without test rendering first. That is time consuming even at 320 x 200.
http://flickr.com/photos/njeneb/
Forget Tuesday; It's just Monday spelled with a T

TheBadger

I did not see any light or shadow poping/flicker, so thats real good!

For me the best part was when the camera flew close to the trees. Lots of nice aspects to the render there.

I'm really interested in your render times tough. They seem way to high! Am I wrong?
It has been eaten.

Bjur

It´s looking goood, m8! :)

I have a few questions regarding to your animation and animation settings generally..

- Were 82 atmo samples really needed if all is raytraced? If yes, why or because of what reasons?

- Max bucket size: I tested different sizes a few times (no animation - Fill light setup, w/o GI) and i ended up with the standart 256 value for best average performance. Was 128 as bucket size rly working faster for you (over the time of some animation test-renderings for example or because 128 bucket size can handle GI renders better/faster)?

- Quality Detail: I love to use 0.X5 values too. Is it just me or can someone confirm that such a value (straight at value ".X5") is looking more like the next higher value without to steal much more render time? Err.., for example: 0.64 is looking more like 0.6 in detail, but right at 0.65, all is looking more like 0.7 detail for me..

- May i ask you: What was your reason or benefit to go over ~AA 6 for motion blurred animations at high resolutions?
In some tests of mine high AA values were the reason my little,  normally sharp looking picture, started to look kinda soft or muddy again. Beside the master of horror, the raytraced shadow itself, extra-high AA settings producing a lot of additional render time, especially in high resolution animation scenes i would bet..

- Ray detail region padding and even more additional render time: Does it help in "fuzzy" landscapes or animations if the value is higher than the standart at max. 1?

Never have worked with any plants till yet i have to admit. That could explain some things too maybe..

I am asking all the stuff because there is a chance of facing animation thingies soon. In the forum there is just a wild mix of opinions, hints, workarounds and results out of all ages..  ;D


Alex
~ The annoying popularity of Vue brought me here.. ~

rcallicotte

OMG!  Very long for such a short beautiful strip of animation.  Someday, there needs to be a better way to do animation without the horsepower.  Not saying this as a criticism of TG2...just don't have the horsepower and presently won't pay someone else to render it (my cheapness   ;D ). 

Glad you were able to do it, though. 

Your mountain ranges need more details...not as realistic as they could be.  And the haze is a little too much...maybe that could explain the long render times, too.  Not sure.
So this is Disney World.  Can we live here?

Tangled-Universe

Quote from: Bjur on November 22, 2012, 07:29:31 PM
It´s looking goood, m8! :)

I have a few questions regarding to your animation and animation settings generally..

- Were 82 atmo samples really needed if all is raytraced? If yes, why or because of what reasons?

Is it ray traced atmosphere? I didn't see that in his post. If so, then this is quite a mistake. If you were to render with AA16 and ray traced atmosphere than just about 2 or 4 samples is really enough.

Even without ray traced atmosphere 82 samples is pretty high. 32 or 48 samples is probably enough.

Quote
- Max bucket size: I tested different sizes a few times (no animation - Fill light setup, w/o GI) and i ended up with the standart 256 value for best average performance. Was 128 as bucket size rly working faster for you (over the time of some animation test-renderings for example or because 128 bucket size can handle GI renders better/faster)?

In my experience, and some others I believe, 128x128 is a bit faster.
You may be right though, that on average (many different type of scenes) it doesn't make that much of a difference.

Quote
- Quality Detail: I love to use 0.X5 values too. Is it just me or can someone confirm that such a value (straight at value ".X5") is looking more like the next higher value without to steal much more render time? Err.., for example: 0.64 is looking more like 0.6 in detail, but right at 0.65, all is looking more like 0.7 detail for me..

Quote
- May i ask you: What was your reason or benefit to go over ~AA 6 for motion blurred animations at high resolutions?
In some tests of mine high AA values were the reason my little,  normally sharp looking picture, started to look kinda soft or muddy again. Beside the master of horror, the raytraced shadow itself, extra-high AA settings producing a lot of additional render time, especially in high resolution animation scenes i would bet..

AA16 can make things a lot smoother, but it also adds a lot of rendertime. It's a matter of determining how smooth you want it to have.
The smoothness of the result is determined by the pixel noise threshold. The lower the threshold the smoother the result *given that there are enough AA samples available to reach that result*(!!!).

This is quite complicated matter, but coincidentially I discussed this on Facebook with a couple of other members and I'll post it here:

What determines the final detail and noise = pixel noise threshold. This determines how smooth the result is.
The lower this number is the more AA samples will be applied (IF available!) and the smoother/cleaner the result.

With pixel noise threshold @ 0.15 the AA algorithm will stop @ 0.15, which is pretty noisy.
For instance: AA8 defaults with a pixel noise threshold @ 0.0375.
Thus 0.15 can *never* give better results than 0.0375.

The lower the threshold the better the result, IF there are enough AA samples available determined by the AA setting.

With AA16 @ 1/64th you will use a minimum of (16x16)/64 = 4 AA samples per pixel.
1/16 = 16 AA samples per pixel.

The lower the threshold the more samples you need.
What you need to find is which AA# (as a power of 2, ideally) samples down to the threshold the quickest.

The tricky part is that some parts of the scene don't need a lot of AA, so you should keep the minimum samples as low as possible, but if you keep it too low the algorithm needs to repeat the sampling too many times to achieve the threshold level and thus it takes longer. You would need to either increase the AA# if you use 1/4th sampling or only switch from 1/64th to 1/16th or from 1/16th to 1/4th sampling.
From 1/4th to full sampling can work, but there is chance that some parts of the scene needed more samples and those would be missed. Thus it is better to increase AA then.

I hope it is a bit clearer now.


Well...it wasn't really....so:

What the pixel noise threshold does is stopping the AA application proces quicker / avoiding that if you use AA16 that all available 256 samples aren't applied, but stops as soon as the threshold reaches 0.15.

The higher the minimum samples the quicker you reach the threshold, but you may oversample parts that way and thus need lower AA or more adaptive sampling (1/16th or 1/64th).


And then one was lost and the other understood it a little, so I hope it is clear to you :)

Quote
- Ray detail region padding and even more additional render time: Does it help in "fuzzy" landscapes or animations if the value is higher than the standart at max. 1?

This requires understanding of what ray detail region padding does, of course.
Here are 2 examples of when you would like to use ray detail region padding:
1) If you have water and out of frame geometry casting reflections into your water.
2) If you have shadows from trees or other geometry which are out of frame but are casting into your frame.

The number for ray detail region padding is a fraction of the amount you expand the camera frustum (frame, so to say).
1 means 100%, 2 means 200% and 0.5 means 50%.

Imagine your camera frame/frustum as a tile, then:
With ray detail region padding @ 1 you will create 9 tiles of ray detail and the centre one is your camera.
With ray detail region padding @ 2 you will create 25 tiles of ray detail and the centre one is your camera.

Why?
A setting of 1 extends the frustum 100% in all directions. So 1 extra at top, bottom, sides and bottom sides = 3x horizontal + 3x vertical = 9
A setting of 2 extends the frustum 200% in all directions. So 2 extra at top, bottom, sides and bottom sides = 5x horizontal + 5x vertical = 25

Briefly summarized:

A ray detail region padding of 2 is seldomly necessary, even 1 is rarely necessary. 0.5 suffices many times.

Quote
Never have worked with any plants till yet i have to admit. That could explain some things too maybe..

I am asking all the stuff because there is a chance of facing animation thingies soon. In the forum there is just a wild mix of opinions, hints, workarounds and results out of all ages..  ;D

Alex

Definitely true there's a wild mix of opinions and such. So is mine an opinion.
Although it may be based on extensive experience, it's definitely not *the truth*.
Only Matt can confirm the sense and nonsense of the opinions here, but he's painfully absent lately.

Oshyan knows a great great deal of animation and he's *very* helpful.

My suggestion is: go animate! start a thread here and ask for advice! :D animate your camera, pick a couple of frames from the whole sequence and experiment with render settings. Try to find rendersettings where the total rendertime is the lowest for all your chosen frames and base those rendersettings on your eye.

Bjur

Thank you for your detailed and enlightening answeres Martin!

Also good to hear about that a 0.5 value for Ray detail region padding seems to be high enough in most cases/ for most scenes.

If possible (moneywise) i will purchase a student licence in the next 1-2 weeks. Then i will try a jump into the deep, cold waters of TG (test-)animations - with all resolutions, AA settings and other unrestricted stuff i wish and need..  :)

~ The annoying popularity of Vue brought me here.. ~

Matt

Quote from: Tangled-Universe on November 25, 2012, 10:01:48 AM
This requires understanding of what ray detail region padding does, of course.

The number for ray detail region padding is a fraction of the amount you expand the camera frustum (frame, so to say).
1 means 100%, 2 means 200% and 0.5 means 50%.

Imagine your camera frame/frustum as a tile, then:
With ray detail region padding @ 1 you will create 9 tiles of ray detail and the centre one is your camera.
With ray detail region padding @ 2 you will create 25 tiles of ray detail and the centre one is your camera.

Why?
A setting of 1 extends the frustum 100% in all directions. So 1 extra at top, bottom, sides and bottom sides = 3x horizontal + 3x vertical = 9
A setting of 2 extends the frustum 200% in all directions. So 2 extra at top, bottom, sides and bottom sides = 5x horizontal + 5x vertical = 25

Briefly summarized:

A ray detail region padding of 2 is seldomly necessary, even 1 is rarely necessary. 0.5 suffices many times.

That's correct, and a good way of explaining it I think.

Let me make small correction to the 2 examples you gave. "Ray detail region padding" only affects shadows cast by terrain or other displaced geometry. Trees and other non-displaced geometry will cast shadows perfectly without needing any padding.

Matt
Just because milk is white doesn't mean that clouds are made of milk.

Tangled-Universe

A thanks Matt, I'll try to memorize that.

Why is that difference actually?

Matt

#11
Undisplaced geometry is kept permanently in memory and can be ray traced immediately. On the other hand, displacements need to be calculated. This takes time and memory. Displacements are stored in the subdiv cache. We need to impose some limits on where the displacement is calculated in order to keep render times and memory use manageable. So we define a "ray detail region". The default region is whatever is visible to the camera or crop region. The padding parameter allows you to expand that if it becomes necessary.

It is called the ray detail region because this is where the highest subdivision detail is calculated to cast shadows/reflections. Geometry outside of this region does still cast shadows/reflections, but it won't have the same level of subdivision as geometry within the ray detail region.

The actual level of detail inside this region depends on the renderer "detail" and the "ray detail multiplier" in the Render Subdiv Settings.

Matt
Just because milk is white doesn't mean that clouds are made of milk.

TheBadger

I like these threads. You guys are talking about things you have explained lots already. But each time its a little different, and it helps me to really grow my understanding.
It has been eaten.

Tangled-Universe

Quote from: Matt on November 27, 2012, 05:18:14 PM
Undisplaced geometry is kept permanently in memory and can be ray traced immediately. On the other hand, displacements need to be calculated. This takes time and memory. Displacements are stored in the subdiv cache. We need to impose some limits on where the displacement is calculated in order to keep render times and memory use manageable. So we define a "ray detail region". The default region is whatever is visible to the camera or crop region. The padding parameter allows you to expand that if it becomes necessary.

It is called the ray detail region because this is where the highest subdivision detail is calculated to cast shadows/reflections. Geometry outside of this region does still cast shadows/reflections, but it won't have the same level of subdivision as geometry within the ray detail region.

The actual level of detail inside this region depends on the renderer "detail" and the "ray detail multiplier" in the Render Subdiv Settings.

Matt

Thanks Matt :)
Ok, so just to check if I understand correctly:

The geometry outside the ray detail region is *always* undisplaced? (the part of your reply in Italic font is the part which made me ask this question)
If these cast reflections into your camera frustum then you get these typical rectangular artefacts.

The geometry in camera frustum is displaced and for example with detail 0.5 and a "ray detail multiplier" at default 0.25, my reflections will be based on subdivisions at 0.125 detail.
Is that correct?

Matt

The parts outside the ray detail region are displaced; it is only the size of the polygons / subdivision level that is different. You can visualise these polygons directly if you set ray detail region to "No detail" and render an image with "ray trace everything".

Matt
Just because milk is white doesn't mean that clouds are made of milk.