Author Topic: Tonemapping - Plz Halp  (Read 115763 times)

2020-05-03, 21:35:08
Reply #135

lupaz

  • Active Users
  • **
  • Posts: 951
    • View Profile

2020-05-03, 21:38:54
Reply #136

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Hmmm.... To me number 8 & 9 are super dull - clamped highlights and grey'ish.

I think nr. 10 definately looks better and more (a)live.

Are we talking taste and style?

They are dull, yes... but the proportions in the contrasts and colors look like in photography! They look correct, as one would expect it in a photo.
Cranking up contrasts or highlights according to your taste shouldn't be a problem... especially when images (like this one) have so much (optical) information in it.



2020-05-03, 21:55:28
Reply #137

lupaz

  • Active Users
  • **
  • Posts: 951
    • View Profile
And here there's a comparison of the same scene, just for the fun of the experiment. Unfortunately the lighting is slightly different, even if the HDRI is the same.




2020-05-03, 22:14:36
Reply #138

twoheads

  • Active Users
  • **
  • Posts: 261
    • View Profile
I know few professional photographers and they use PS extensively. I'm not talking about developing process but pure postproduction. They're like michelin star chefs with butter. They put tons of this s..t everywhere. It takes a lot of time to make photo look great for publication believe me.

My reply about "which render looks better"  wasn't of course 100% serious. Now Imagine fstorm dev watching this topic laughing at us. "what the hell are they trying to achieve here?" :)

As for westkaai, we are discussing work we all know quite well (most of us) and I don't think we are talking about realism exclusively (it's not possible at this point). It would be fair to test both engines in some kind of blind test. Let's say 30 selected images we all have never seen before:  10 for corona, 10 for fstorm and 10 real photos and then we could debate for days and days and finally get conclusion "fstorm is 18% more photorealistic" who cares?

I'd really like to know what devs think about improvements we are talking about, I don't see any waving hands...

2020-05-03, 23:38:06
Reply #139

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
I know few professional photographers and they use PS extensively. I'm not talking about developing process but pure postproduction. They're like michelin star chefs with butter. They put tons of this s..t everywhere. It takes a lot of time to make photo look great for publication believe me.

My reply about "which render looks better"  wasn't of course 100% serious. Now Imagine fstorm dev watching this topic laughing at us. "what the hell are they trying to achieve here?" :)

As for westkaai, we are discussing work we all know quite well (most of us) and I don't think we are talking about realism exclusively (it's not possible at this point). It would be fair to test both engines in some kind of blind test. Let's say 30 selected images we all have never seen before:  10 for corona, 10 for fstorm and 10 real photos and then we could debate for days and days and finally get conclusion "fstorm is 18% more photorealistic" who cares?

I'd really like to know what devs think about improvements we are talking about, I don't see any waving hands...



Well, me for example, I do care... :)
And however... in my opinion, fiddling for hours in PS on an image cannot be the way to go.
3D software devs cannot expect that from users, nor do our clients want to wait for ages or pay a fortune for a single pic.
« Last Edit: 2020-05-03, 23:43:26 by Designerman77 »

2020-05-03, 23:39:52
Reply #140

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
And here there's a comparison of the same scene, just for the fun of the experiment. Unfortunately the lighting is slightly different, even if the HDRI is the same.






Thumbs up.
As I mentioned in some post before... you often see the better tone mapping once you desaturate an image to BW.
Cool test you did again!

Clearly visible differences...
« Last Edit: 2020-05-03, 23:44:10 by Designerman77 »

2020-05-03, 23:55:41
Reply #141

Juraj

  • Active Users
  • **
  • Posts: 4743
    • View Profile
    • studio website
I would love to see the test above continued if you'll find time. Isn't the difference coming from mirrored HDRi? Every single loader, every engine, has this standardized differently (including starting angle!).

The region, top left corner is the most interesting difference.
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-05-04, 00:31:19
Reply #142

cjwidd

  • Active Users
  • **
  • Posts: 1077
    • View Profile
    • Artstation

2020-05-04, 02:28:04
Reply #143

lupaz

  • Active Users
  • **
  • Posts: 951
    • View Profile
I'm keeping noise in the images so we don't pay too much attention to shaders.

They're pretty much the same in my opinion.

What it is important, I think, and may be the reason why people think the problem is tone mapping, is that the shadows on the corners are stronger in Fstorm, even when the overall image has less contrast.
With the corona image, even after applying a strong post production, the corners will not have that nice shadow that you have in the fstorm image. May be using AO?
So I agree with the person that said that highlights and shadows are stronger in fstorm.

Also I notice that the geometry gets defined better (or sooner?) in fstorm. Look at the knob on the door, or the charger in the post on the counter

Just a note: I'm not saying that Fstorm is a better engine than Corona. Just comparing.





2020-05-04, 04:05:35
Reply #144

lolec

  • Active Users
  • **
  • Posts: 179
    • View Profile
I notice you didn't enable LUT or modify Hightlight compression. Any reason why? Fstorm applies a LUT by default, so this is not a oranges to oranges compression, just wondering if you have a reason?

2020-05-04, 07:38:06
Reply #145

Fluss

  • Active Users
  • **
  • Posts: 553
    • View Profile
Every single loader, every engine, has this standardized differently (including starting angle!).

This what hurt the most in those comparisons... Fstorm loader is indeed mirrored.. It seems like lupaz corrected it in the last render, not saying much about it.. Guys, checking the technical side is crucial in those kinds of comparisons, so as explaining the process... Everybody was taking a conclusion on renders with different lighting.... That sucks.

A lot of posts above were complete garbage.. comparing real photos to renders with completely different subjects, no controlled environment etc etc.. No need to expand more, this is useless. Let's not make this thread a complete junk pile.

Please lupaz, keep conducting your current test, it's interesting. But please make it fair and accurate. First thing would be to make them 1:1. So no missing materials/maps (if something do not work in an engine, should be changed by something that both supports in both scenes), same lighting (seems to be a little bit off still, based on the highlights upstairs), and a matching exposure. This would be a good starting point on which you can iterate.

Also, Fstorm as no caching solution, it's BF+BF. Not saying you have to change this one but this might explain some really fine details discrepancy some are talking about. Fstorm was remapping the glossiness range from [0.4;1.0] to [0.0-1.0], is it still the case? This could have a huge impact so keep that in mind.
« Last Edit: 2020-05-04, 07:53:39 by Fluss »

2020-05-04, 08:01:06
Reply #146

cjwidd

  • Active Users
  • **
  • Posts: 1077
    • View Profile
    • Artstation
^thank you @Fluss

2020-05-04, 10:21:19
Reply #147

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Every single loader, every engine, has this standardized differently (including starting angle!).

This what hurt the most in those comparisons... Fstorm loader is indeed mirrored.. It seems like lupaz corrected it in the last render, not saying much about it.. Guys, checking the technical side is crucial in those kinds of comparisons, so as explaining the process... Everybody was taking a conclusion on renders with different lighting.... That sucks.

A lot of posts above were complete garbage.. comparing real photos to renders with completely different subjects, no controlled environment etc etc.. No need to expand more, this is useless. Let's not make this thread a complete junk pile.

Please lupaz, keep conducting your current test, it's interesting. But please make it fair and accurate. First thing would be to make them 1:1. So no missing materials/maps (if something do not work in an engine, should be changed by something that both supports in both scenes), same lighting (seems to be a little bit off still, based on the highlights upstairs), and a matching exposure. This would be a good starting point on which you can iterate.

Also, Fstorm as no caching solution, it's BF+BF. Not saying you have to change this one but this might explain some really fine details discrepancy some are talking about. Fstorm was remapping the glossiness range from [0.4;1.0] to [0.0-1.0], is it still the case? This could have a huge impact so keep that in mind.


May I ask what exactly is your point? You claim that Corona is a perfect render engine and devs. could go home and stop optimizing it?
It is NOT perfect and still needs a good amount of work until it can be called really photo-realistic. PERIOD !

Calling other peoples` thoughts "garbage" is an offense to all the guys here that take their precious time, energy and experience, willing to
make this great software even better.

I am sure the devs at Corona are not offended that users make suggestions. Actually, the nowadays quality of this software is based on the open minded attitude of the Corona dev team, not on an attitude of ignorance and "defending the existing limitations".

2020-05-04, 10:45:59
Reply #148

jms.lwly

  • Active Users
  • **
  • Posts: 171
    • View Profile
    • jms.lwly studio
it is absolute guesswork every time

Revisiting cjwidd's original point (things seem to have deviated from his original question - although brought about an interesting new conversation...) - every project I have the same back and forth in my setup:-

Use a LUT? Don't use a LUT, just use HC and Filmic...
Do less in VFB and more in Photoshop? Try to squeeze the very best output from VFB...
Use just HDRI for environment lighting? Use Corona Sun/Sky for lighting...

I realise with the all of above there is no correct answer - but I think this was the original question, which I fundamentally struggle with too.

2020-05-04, 10:54:00
Reply #149

Fluss

  • Active Users
  • **
  • Posts: 553
    • View Profile
May I ask what exactly is your point? You claim that Corona is a perfect render engine and devs. could go home and stop optimizing it?
It is NOT perfect and still needs a good amount of work until it can be called really photo-realistic. PERIOD !

Calling other peoples` thoughts "garbage" is an offense to all the guys here that take their precious time, energy and experience, willing to
make this great software even better.

I am sure the devs at Corona are not offended that users make suggestions. Actually, the nowadays quality of this software is based on the open minded attitude of the Corona dev team, not on an attitude of ignorance and "defending the existing limitations".

I'm just asking for a pragmatic approach. If you follow my post history, you'll actually see that I'm trying to push for some improvements as well. But come on, comparing a photograph of a kitchen with a render of a completely different scene, Making a poll with random renders... what's the point of this? This is useless as this proves nothing -> garbage. I'm not defending anything, you do.

I'm sorry for you but @BardhylM was right.

disclaimer : just to clarify, when I said no controlled environment, I was talking about the process of comparing real photos to renders. That part was not about lupaz test which is great even if it has some flaws I pointed out. So your test lupaz was not included in the garbage stuff.
« Last Edit: 2020-05-04, 11:27:49 by Fluss »