Author Topic: Corona 1.x lost the most important core algorithms  (Read 9159 times)

2016-02-11, 20:16:18

M.S.Referee

  • Active Users
  • **
  • Posts: 11
    • View Profile
I'm missing Corona Render alpha v7.2 very very much, because the algorithms, especially the "Bucket" algorithms.

Yep, alpha v7.2 is slow than v1.x, but alpha v7.2 is the real Corona Render.

Slow of alpha v7.2 is built on the premise of guaranteeing quality of this unique algorithm, it's just need to improve efficiency of the implementation of this algorithm and add new features like v1.x did.

Although v1.x is that seems to be faster than alpha v7.2, and v1.x got good new features, but v1.x lost the most important core algorithms which alpha v7.2 got, and v1.x slow than alpha v7.2 when over
.
.
.
The algorithm, alpha v7.2 is "Checkerboard Algorithm" which I feel that, and this is the core of Corona Render, and this is why you can clear to get very good result just from few mins and few passes in "Bucket" mode.  For example in "Bucket" mode, if you run 10 pass with only 2 samples for each pass, even if this is pretty much get on passes and very good result.
But, in v1.x, even if you run 75 pass, it's like a joke, you still can easy saw noise.

So, algorithm of v1.x is a "Fake Fast" which I called, it's no built on the premise of guaranteeing quality of this unique algorithm, just blindly seek for quick with abandoned core algorithm. That's so bad! Even if when you have to use lots of passes to get good result like alpha v7.2 did in v1.x, in fact, it's already slower than alpha v7.2. So it's a change of more harm than good.

Just like the antialiasing of alpha v7.2 is definitely good than v1.x, especially for hypotenuse edge, alpha v7.2 can easy take great result of antialiasing hypotenuse edge(sure, including other type of edges) with few pass. But v1.x can't. (Both use 'Tent' to get result)
.
.
.
So, I real hope your team back to the real foundation algorithm of alpha v7.2 for Corona Render and bring those new features from v1.x and improvement to this up of foundation, that's the real right way for formidable Corona Render, otherwise Corona Render will lose the biggest advantage forever just like the v1.x now.
.
.
.
And a suggest for feature, you can make GPU to do denoise when rendering, that's maybe a new way to save speed of rendering and not affect quality.
.
.
.
Thanks very much your team's hard work and made the real great render!!!
.
.
.
.
.
.
1.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Forest Pack 4, Sample 18, Passes 1, Primitives 238M, Prepare Scene Time: 3 secends, Rendering Time 02:55. You can directly do denoise for got final picture.

.
.
2.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Forest Pack 4, Sample 8, Passes 1, Primitives 70,950,495, Prepare Scene Time: 1 secends, Rendering Time 01:54. You can directly do denoise for got final picture.

.
.
3.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Forest Pack 4, Sample 18, Passes 1, Primitives 68,246M, Prepare Scene Time: 42 secends, Rendering Time 05:32. You can directly do denoise for got final picture.

.
.
4.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Forest Pack 4, Sample 8, Passes 1, Primitives 68,246M, Prepare Scene Time: 42 secends, Rendering Time 02:07. You can directly do denoise for got final picture.

.
.
5.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Laubwerks Plants Kits, Sample 1, Passes 3, Prepare Scene Time: 01:42 secends, Rendering Time 10:57. You can directly do denoise for got final picture.

.
.
6.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Forest Pack 4, Sample 1, Passes 2, Prepare Scene Time: 6 secends, Rendering Time 11:06. You can directly do denoise for got final picture.

.
.
7.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Sample 1, Passes 3, 792 Copy Objects more than 81,000,000 polys, Prepare Scene Time: 4 min 13 secends, Rendering Time 01:40. You can directly do denoise for got final picture.
And this scene render with Vray 3, prepare scene need more than 30 mins, prepare GI need more than 1 hour, final rendering need about 7 mins.

.
.
8.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1280x720, Sample 6, Passes 4, Average Prepare Scene Time: 0.1 secends, Average Rendering Time: 30 secends. You can directly do denoise for got final picture. This is a animation which got 100 frames. And finished rendering just need 50 mins.

.
.
9.One Xeon E5 2670v1, 16GB DDR3 1333HZ Memory Single Channel, Corona Render Alpha v7.2, Bucket Mode, 1920x1080, Forest Pack 4, Sample 1, Passes 4, Primitives 3,864M, Average Prepare Scene Time: 2 secends, Average Rendering Time 00:55 - 01:25. You can directly do denoise for got final picture. This is a animation which got 240 frames. And finished rendering just need 3 hours.
« Last Edit: 2016-02-12, 22:23:10 by M.S.Referee »

2016-02-12, 09:04:19
Reply #1

tomasd

  • Active Users
  • **
  • Posts: 19
    • View Profile
Can you please give us a screenshot of what exactly you mean by the "Checkerboard Algorithm."

2016-02-12, 13:16:30
Reply #2

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 11277
  • Marcin
    • View Profile
I have no idea what this thread is about. :/
We will need some more explanations - example images would be great.
I can only say that I am sure Corona gets better with each new version. If one is experiencing some problems with a new version, which were not present in the older version, then my guess would be that it's either user error, or some compatibility issue between the two versions which should be easily fixable, for example by using the converter script.

There is bucket rendering in the newest version of Corona and it is working almost the same way as in A7.2 - only the controls are more intuitive now (in older versions the number of samples would be squared or multiplied by 4 (I could never remember this) - and now simply the specified number of samples is added per each pass). It should be possible to achieve the same results with A7.2 buckets and newest buckets.
Marcin Miodek | chaos-corona.com
Chaos Corona Support Team Lead | contact us

2016-02-12, 14:17:07
Reply #3

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 9042
  • Turning coffee to features since 2009
    • View Profile
Can you please give us a screenshot of what exactly you mean by the "Checkerboard Algorithm."
The Checkerboard algorithm is silently disabled in cracked version of corona
Rendering is magic.
Private scene uploader | How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2016-02-12, 14:51:22
Reply #4

arqrenderz

  • Active Users
  • **
  • Posts: 893
  • arqrenderz.com
    • View Profile
    • arqrenderz

2016-02-12, 14:58:36
Reply #5

FrostKiwi

  • Active Users
  • **
  • Posts: 686
    • View Profile
I was already curious to ask, whether there was some smart detail included like V-ray's camera shake'n'flicker.

2016-02-12, 17:01:26
Reply #6

M.S.Referee

  • Active Users
  • **
  • Posts: 11
    • View Profile
Hi, maru

Your team can do a test between alpha 7.2 and v1.x with Bucket Mode. Because it will very easy to know different and what is I called that "Checkerboard Algorithm".

Test scene type,
1). Medium shot with pure white diffuse material and Corona Sun.
2). Medium shot with pure reflection material and Corona Sun.
3). Medium shot with pure refraction material and Corona Sun.
4). Medium shot with mix those 3 type of material and Corona Sun.
Settings,
1). Turn to Bucket Mode for all test scene.
2). Set sample to 64 for pure white diffuse material, set sample to 16 for pure reflection material, set sample to 24 for pure refraction material.
3). Set pass to just 1 for all test scene.
4). Multi-pass, set sample to 1-4 and 5 pass, then do the four test for each sample setting.

Then rendering, and check look the quality and speed between alpha 7.2 and v1.x.
You will got clear feeling:
1) In alpha 7.2, the speed is true very slow with 64 sample or high, but the result is very very good, almost can't see noise. And you can check look the hypotenuse edge of antialiasing, it' so great. You still can see reflection and refraction got very good result and the speed is fast too. And Multi-pass test, the result sure good than v1.x and speed slow than v1.x.

2) In v1.x, the speed is absolutely fast than alpha 7.2, but the result never can be compared with alpha 7.2, you can clear see noise. And you can check look the hypotenuse edge of antialiasing too, it's OK.

Conclusion:
This is why I'm so missing alpha v7.2 of Corona Render, because when work with high sample of Bucket Mode, and if you want to render animations, especially with render farm, you just need to do 1 pass or at most 2 passes with 64 sample with alpha v7.2, you will got the final result. Even if you don't need to do denoise.
And for some type of material(like more reflection, pure refraction) scene, you just need to adjust sample with only 1 pass, maybe 1 min or 2, 3 mins, you will got final result.
But with v1.x of Corona Render, maybe you need ten times or series ten times passes than alpha v7.2 with same sample to got final result. Maybe you still need to do denoise.

So I don't think v1.x good than alpha v7.2 about algorithm. And the "Checkerboard algorithm" from Corona first verion to Corona alpha v7.2, and gone from Corona Daily v1.0 to now.
« Last Edit: 2016-02-12, 18:06:12 by M.S.Referee »

2016-02-12, 17:11:35
Reply #7

M.S.Referee

  • Active Users
  • **
  • Posts: 11
    • View Profile
I was already curious to ask, whether there was some smart detail included like V-ray's camera shake'n'flicker.

About flicker of animation, v1.x got no problem with UHD Cache and Path Tracing, but alpha v7.2 got this serious problem with HD Cache, and Path Tracing better than HD Cache.

2016-02-12, 17:32:02
Reply #8

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 11277
  • Marcin
    • View Profile
1. It makes completely no sense to set bucket passes to 1 in A7.2 or any other version of Corona. You will get EXACTLY the same result as when using progressive mode and 64 passes. If bucket passes is set to 1 - then no adaptivity will be used, so there will be no advantage over progressive rendering whatsoever.

2. Can you post any examples that would confirm your thesis?
Marcin Miodek | chaos-corona.com
Chaos Corona Support Team Lead | contact us

2016-02-12, 18:38:28
Reply #9

M.S.Referee

  • Active Users
  • **
  • Posts: 11
    • View Profile
1. It makes completely no sense to set bucket passes to 1 in A7.2 or any other version of Corona. You will get EXACTLY the same result as when using progressive mode and 64 passes. If bucket passes is set to 1 - then no adaptivity will be used, so there will be no advantage over progressive rendering whatsoever.

2. Can you post any examples that would confirm your thesis?

I'm sure that Bucket pass 1 is real meaningful with different samples in alpha v7.2, and it's no different 64 sample with 1 pass and 64 pass with 1 sample in v1.x, because this is not my imagination, this is the result which I have personally tested.

Yes, sure, I saved some test.
« Last Edit: 2016-02-12, 19:46:30 by M.S.Referee »

2016-02-14, 07:03:34
Reply #10

tomasd

  • Active Users
  • **
  • Posts: 19
    • View Profile
Instead of comparing the specific settings, can you compare time to same quality?

2016-03-27, 14:45:20
Reply #11

rafpug

  • Active Users
  • **
  • Posts: 723
    • View Profile
Hello M.S.Referee
and Happy Easter everyone

for just curiosity:
You can compare two examples , render times and image quality from vers. 7.2 to the latest update of Corona Render ?
..and the temperature of the processor between the two versions ?



It ' still possible this comparison ?

Thx
Rafpug

2016-03-27, 14:48:50
Reply #12

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 9042
  • Turning coffee to features since 2009
    • View Profile
He can't, he is currently banned for using crack :D
Rendering is magic.
Private scene uploader | How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2016-03-27, 14:54:02
Reply #13

rafpug

  • Active Users
  • **
  • Posts: 723
    • View Profile
Hello Ondra

It 's always a pleasure to read you , your team is doing a great job