Author Topic: Animation - losing frames  (Read 6224 times)

2020-08-13, 00:41:49
Reply #15

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
One more question, Tom:

at noise level 4% I find many surfaces - such as building walls - to still show too much noise. In the animated camera movement it gets much more visible as flickering
compared to still images.

Denoiser is already at 0.8

UHD is cached. So I would expect even less visible noise & flickering.
Yep, "flicker free" is also on.

Going lower with the noise level to 2% or even lower is no option for me...in terms of render time going exponential. And I find in general this is e very extreme step.
Not everyone has set up a massive render farm.

2020-08-13, 00:46:29
Reply #16

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
Do those walls have bump on them? The Corona HQ Denoiser does less denoising where there's bump (to preserve detail) so that could be the reason, would be interesting to see some frames. Various things you could try, such as, a VirtualBeauty with even higher denoising and a mask for the walls (eg Denoising of 1) and compositing that super denoised version in post; or you could try one of the AI denoisers, which don't know about the presence or absence of bump in the scene so denoise equally.

Of course it might also not be rendering noise, it could be other things, such as just the bump catching the light differently from different camera angles, or displacement set to screen space rather than world space and changing resolution as cameras move. Again, impossible to say what kind of flickering/noise is present without seeing the result.

Also, is Sharpen/Blur used? That can help, if not.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2020-08-13, 02:58:22
Reply #17

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Do those walls have bump on them? The Corona HQ Denoiser does less denoising where there's bump (to preserve detail) so that could be the reason, would be interesting to see some frames. Various things you could try, such as, a VirtualBeauty with even higher denoising and a mask for the walls (eg Denoising of 1) and compositing that super denoised version in post; or you could try one of the AI denoisers, which don't know about the presence or absence of bump in the scene so denoise equally.

Of course it might also not be rendering noise, it could be other things, such as just the bump catching the light differently from different camera angles, or displacement set to screen space rather than world space and changing resolution as cameras move. Again, impossible to say what kind of flickering/noise is present without seeing the result.

Also, is Sharpen/Blur used? That can help, if not.


Hey Tom,

thanks a lot for your reply & suggestions.

AI denoiser is already used in the scene for reasons you mentioned. I actually always use the AI denoiser, since it doesn't mess up bump maps & round edges.

Yes, a little bit of bump is used on the walls. But from that big distance, it actually should not have a visible effect... in my understanding, at least.
But who knows.

You mentioned UHD caching with the "load & append" method. What is the advantage compared to simply "load from file"-method?


Thanks in advance for getting back.





2020-08-13, 03:05:01
Reply #18

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
Well, you would use Load + Append while generating the saved version of the UHD Cache - this basically calculates the cache for one camera location, then loads that and calculates the cache for another location adding that result to what was calculated before, and then repeat as it goes through the animation.

Then you simply Load when rendering, as the cache has information from various camera angles added in to it. A good example is to think of a camera starting in one room, and moving through a doorway into another room - if you only calculate the cache from the first room, nothing is calculated for the second. With load and append in the calculation phase, cache information for both rooms is generated.

(You don't really want to use load and append during rendering, as then you are pretty much just calculating the UHD Cache each time and would be as well calculating from scratch - so it's during the generation phase you want to use it).
Tom Grimes | chaos-corona.com
Product Manager | contact us

2020-08-13, 03:29:55
Reply #19

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Hey Tom, thanks for the explanation... despite it's late at night. :))))

"Load & append" takes all camera positions of an animation into consideration, from how I understand you?
Which should actually be a big advantage... ( that may even eliminate that flickering occurring due to every new frame / camera position change).

So, "load&append" is chosen only while generating & saving the UHD cache. After finishing the calculation of the cache, one switches back to "load from file" during the rendering process.

2020-08-13, 03:36:44
Reply #20

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
Right - Load and Append only during generating the cache (where you don't have to have more than 1 pass, and can do it every 5 or 10 or 30 frames, depending on how much things are changing). Then just Load during rendering.

And yep, accounts for all camera positions. Basically calculates it and saves it, loads what it saved, calculates it from the new position and adds that in to what's there already and saves that, loads that and.... and so on :)

You can test it with a simple scene, e.g. a cube where camera circles around it. Just calculate and save while looking at the front face, when animation moves to the back face, there's no Cache calculated for it. With the Load and Append generate process, then Load during rendering, there's info in the Cache for the front face and for the back face (the Cache now has info for various camera positions stored in it).
Tom Grimes | chaos-corona.com
Product Manager | contact us

2020-08-13, 13:34:11
Reply #21

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Right - Load and Append only during generating the cache (where you don't have to have more than 1 pass, and can do it every 5 or 10 or 30 frames, depending on how much things are changing). Then just Load during rendering.

And yep, accounts for all camera positions. Basically calculates it and saves it, loads what it saved, calculates it from the new position and adds that in to what's there already and saves that, loads that and.... and so on :)

You can test it with a simple scene, e.g. a cube where camera circles around it. Just calculate and save while looking at the front face, when animation moves to the back face, there's no Cache calculated for it. With the Load and Append generate process, then Load during rendering, there's info in the Cache for the front face and for the back face (the Cache now has info for various camera positions stored in it).




Hey Tom,

thanks a lot! All these priceless informations that actually belong in a Corona-how to-manual. :)

Gonna check the difference between the two UHD cache methods.

The only point I don't fully understand is: what do you mean by "every 5 / 10 or 30 frames?
To jump through the animation very X frames and save&append the UHD - which creates several files... and then during the animation to manually change the UHD cache file?
I can't imagine that you meant like this. :)

Sorry if the question might be stupid due to my misunderstanding... but I rather ask stupid stuff instead of doing stupid things in the renders. :)


2020-08-13, 14:07:34
Reply #22

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
Load and Append should write to the same file, not creating multiple files.
Tom Grimes | chaos-corona.com
Product Manager | contact us

2020-08-13, 15:23:10
Reply #23

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Load and Append should write to the same file, not creating multiple files.

Hey Tom.

Yep, that's what I would have expected.

However, I'm doing a test of that same animation at full resolution with the "Load & Append"- caching right now.

Already the first frames show WAY ! less flickering (same settings), compared with the "simple" Load From File - method.


Aaaargh... why have I not known it before this night 200 frames got ready rendered???
Shaaaait. :)))))



UPDATE: after comparing more frames between the two animations... the impression to have "less flickering" with the UHD "load & append", seems to have been just self-suggestion / illusion based on positive expectation.
If there is a difference, it is very minimal.


I`m gonna try to reduce the flickering in After Effects.
« Last Edit: 2020-08-13, 21:36:35 by Designerman77 »

2020-08-14, 00:07:15
Reply #24

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Best solution against flickering: render in much bigger resolution ( e.g. 2x ) and leave the pass number roughly the same.
Quality will be good... and even better if you are going to reduce the size when editing the footage.

Plus a tip I read from Maru: lowering the GI/AA, and applying sharpen/blur.
Gets rid of even more flickering / noise.


No more head ache.