Tonemapping - Plz Halp

[cjwidd]

@lupaz definitely not, but emerging color standards are helping to increase parity between different renderers

[cjwidd]

Cross-posting a thread about calculating Corona EV from 'I need help' because it's relevant to the current discussion:

EV vs. Corona EV

[Fluss]

For such trivial things as exposure, I expected it to be more or less the same. The thing is that if I raise my ISO from 1600 to 12800 ISO, I should have +3EV.
In corona, a render at  1600ISO, +3EV in post looks more or less the same as a render at 12800 ISO. In Vray, at least in V5 beta, it's not the case.
-> In Vray, I was limited by Max sample intensity.

[James Vella]

Is it even possible to match every technical aspect between render engines anyway?

@lupaz I dont think so, nor is it entirely important - what I find important is parity between materials. You should be able to change lighting based on preference and if the materials are setup correctly (PBR for example) then you know you can adjust the lighting/tonemapping to taste without borking something else along the line.

Well, always though Vray and corona photographic exposure settings would match but after quickly looking at it a bit closer, it does not seems to be the case. I'll try to dig a bit further if I find time.

@Fluss I think your first thought was correct, so you are not going crazy :) I found that with some testing this morning 1600 - 12800 is exactly 3 stops. Thanks to @cjwidd and @maru for posting the thread on the vs EV thread I was able to use the excel sheet to do the conversion and it seems correct to me.

Vray 1600 ISO / 0 Exposure



Vray 1600 ISO / +3 Exposure



Vray 12800 ISO / 0 Exposure



Corona 1600 ISO / 0 Exposure



Corona 1600 ISO / +3 Exposure



Corona 12800 ISO / 0 Exposure



Just an FYI if you are trying to follow along in Corona, I had to switch this to a Physical Camera as the Corona Camera was giving me strange results. Not entirely sure why - maybe its just because I always use Physical Camera for vray/corona and used to the setup.

edit:
My conclusion is that I still think the ACES workflow is producing the result that is most similar to photographic workflow (lightroom). Where if you have a well exposed image with no crushed areas then you should be able to push the light either way without compromising your highlights somewhere (actually more so since we are working in 32bit float). Similar to how you can blow out the fstorm render and then pull it back with the burn setting, this is impossible with Corona's highlight compression.

[Fluss]

I was using renderer's native camera in both. We're definitely experiencing different results here.

[James Vella]

Yes it took me a few moments to work out how to get parity between them at 1600 / 0 Exposure, thats because there was some limitations with the Physical Camera. My mistake, I used Vray Camera for Vray, Physical Camera for Corona (as I had issues with the Corona Camera overriding my global settings).

This should get you off on the right foot for the comparison.

Vray Settings (Vray Camera):



Corona Settings (Physical Camera):



3D Files (3dsmax 2019):

https://we.tl/t-K37vLOCyk0

For this setup I had to change the fstop (16) shutter speed (2000) for a good base point for adjusting the exposure 3 stops. This way you should have both on parity from this point onwards. Based on the excel document in Corona thats -2.813 for +3 exposure, and -5.813 for 0 exposure. Vray you just adjust the exposure without any need for a conversion.

[Fluss]

I downloaded your scene and it worked as expected. But if you try with my camera settings (f4, 1/50s, 1600/12800) and load the HDRi I used for the test (Vraymap, spherical mapping, everything else default), it doesn't work anymore. really strange. Is it the case on your side? (Also, no tone mapping, standard sRGB display)
(Note that with this setup, we're closer to image burn)

[Fluss]

Found out that max ray intensity was the "culprit" in Vray. Disabling it solved the issue. I edit my previous messages, it's kind of a mess. Maybe it would be better to delete them

[James Vella]

Found out that max ray intensity was the culprit in Vray. Disabling it solved the issue.

Yep, that did the trick!

[Fluss]

This makes me think the more "directional look" of Fstorm is due to heavy GI clamping.

[Fluss]

Reinstalled Fstorm,  and damn! I don't like it! I forgot how it feels, well, completely different. In a bad way. It seems the glossiness is indeed still mapped from 0.4 to 1.0.

[James Vella]

deleted, reinstalled Fstorm,  and damn! I don't like it! I forgot how it feels, well, completely different. In a bad way. It seems the glossiness is indeed still mapped from 0.4 to 1.0.

Takes a few moments to work some things out. I really like how simple it is, just be aware all bitmaps come in at the correct gamma no need to change anything except HDRI's should be set to 2.2 (kind of opposite but first time user friendly in a way, same goes for render settings in tone mapping leave the gamma settings alone). One thing that I find difficult is there is no camera settings like shutter speed, fstop doesnt affect the lighting like a camera does (that I have found yet, I dont really use it either to be honest just tests every now and then). This will also affect these tests since we cant really bench it against each other. Exposure seems to be a bit arbitrary (or I havent found the recipe for it). But I think this is the charm of it, you dont need to be a technician to use it, you just adjust the exposure/tone mapping and the rest is taken care of without much thought - which is kind of where I see this conversation going, people want an out of the box photo real experience, not having to think about gamma settings or camera settings and get the photo real result. Technical people will probably have a melt down at first but theres probably more to it that Im unaware of, just going by what I read of peoples blog posts etc.

[Fluss]

TBH I'm fed up with those comparisons.. I'm gonna stop there, the more I dig into it, the more I feel it pointless... I always come to the same conclusion, a renderer is a tool, and if used wisely, you can produce beautiful results with any of them.

[James Vella]

lol thats fair enough, thanks for chiming in its been an interesting experiment.

... I always come to the same conclusion, a renderer is a tool, and if used wisely, you can produce beautiful results with any of them.

I completely agree with you.

For the experiments I posted above, I was interested in keeping focus on just one variable - exposure, and disregarding tone mapping entirely for the sake of trying to single out of this variable has any effect on the flexibility of the lighting control from the artists perspective after rendering. From this point onwards introducing one additional variable - highlight compression, and what I expect to be able to control once I have my rendered image with tone mapping.

I once proposed devs what I consider could work as workaround. To allow Corona to save directly into .dng format. Adobe's DNG has both 16bit and 32bit floating point options. And I thought perhaps this would confuse raw editors less, effectively "trick" them.

I think Juraj is on to something here. For people in the archviz field this would be a great option.

For those who work in film, they already work in 32bit space with linear images and thus have a solid pipeline for post production corrections.

Speaking on behalf of myself (and assuming others in this thread), we want that same control in our VFB, we want to render like photographers. Which is why I think fstorm is toted as having excellent tone mapping controls.

edit:
Instead of rendering like a photographer, what I mean to say is render as the human eye perceives things.

[Designerman77]

TBH I'm fed up with those comparisons.. I'm gonna stop there, the more I dig into it, the more I feel it pointless... I always come to the same conclusion, a renderer is a tool, and if used wisely, you can produce beautiful results with any of them.

Not with "any of them". Otherwise there would not be different render engines on the market.

Before I chose Corona as an renderer, I compared ( I guess ) hundreds of images from the net, coming from Octane, VRay, Arnold, etc. etc.
One could clearly see differences. Guess which difference was the most obvious to me? Lighting & color grading!

Some render engines create (in average) more artificial images than others. Corona was and is among the top three engines that have beautiful, realistic lighting and especially a pleasant atmosphere in images.

But as with everything in life, everything is great until you see something greater.
And sorry to say... but from FStorm I see images from at user levels - and it is immediately visible that the engine has a more photoreal way to calculate light, contrasts and colors.
We can make 1 Mio. tests here with trivial little chambers and a ball in the background. :))))