linear gamma in resolve and mlvapp

Started by 70MM13, January 28, 2019, 03:53:29 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

70MM13

I saw someone on here mention using linear gamma in resolve, so I gave it a shot, and IT ROCKS.

I loaded up a video I did last year when I was first playing with reduced gain, and for the first time, I was able to get things looking right.

This video was not shot to be nice looking.  It was a series of tests of extreme conditions to determine the limits of the camera with reduced gain.  The whole thing was shot using iso 200 dropped to approximately 109.  For example, the only light in the entire video is a single kitchen light bulb in the opening segment which was shot before sunrise.  Extremely harsh and bad lighting in a dim area with just enough cold blue light from outside to make it look really ugly.  And yet, it looks pretty good!

The shot of the paddle in the water was in full sun at noon, directly overhead.  This shot, and the melting icicle, puts to rest any false notion that you can't get good highlights with gain reduction.  You have to expose for it, and the clean shadows means you can lift them way up.

It also means you have to do serious grading to do this while looking natural.  Linear gamma did the trick.  It required a very strong inverse log curve, something you could never get away with in any other situation...

There's NO noise reduction used in any of this footage, including chroma.

After I completed this, I loaded up the same footage in mlvapp and made a similar crazy curve in linear mode, and it looks excellent!

The only thing lacking from mlvapp to prevent me from getting the same results is the lack of curves for "lum vs sat" and "sat vs sat".  Those, plus nodes, and mlvapp is ready for anything...

PS: If gain reduction was used for the "dark" iso in dual iso (e.g. 800), it would be even better with absolutely noise-free shadows.  In testing I have done for the last couple of days, I am finding that for some scenes, iso 109 still wins.  Dual iso always wins for extreme dynamic range scenes, so I will be using both from now on as needed.  It would just be nice to have even cleaner "ultra dark" details with dual iso.

vimeo forced me to convert rgb to rec709 so it's not 100% accurate colour, but it's extremely close.


https://vimeo.com/313725903

masc

Quote from: 70MM13 on January 28, 2019, 03:53:29 PM
...
After I completed this, I loaded up the same footage in mlvapp and made a similar crazy curve in linear mode, and it looks excellent!

The only thing lacking from mlvapp to prevent me from getting the same results is the lack of curves for "lum vs sat" and "sat vs sat".  Those, plus nodes, and mlvapp is ready for anything...
...
Interesting. Could you upload a receipt with this crazy curve, or at least a screenshot how it has to look like? I tried that but it won't look better (by far) as normal... I think I am doing it wrong or different to you. Thx.
5D3.113 | EOSM.202

70MM13

Here's an example i already have lying around from another video when I first tried linear gamma in resolve...

The curve is the same for mlvapp.

But...  this won't work for footage that is exposed "normally".  These videos were exposed for the gain reduction mode, so they are extremely dark and i'm lifting the shadows unbelievably high, as you can see from the curve!

It works beautifully in linear gamma.  I wasn't able to get these results in other modes, especially so easily!

Some of the scenes in the vimeo video had fairly "easy" lighting and didn't require such an extreme curve.  That's probably what you need, since you most likely exposed correctly!

:)




masc

Okay, thx. But in MLVApp this looks still strange... you set profile to "linear" and then you paint such a curve? When I set to "linear", 50% looks only black and the few existing highlights have a huge saturation...
5D3.113 | EOSM.202

70MM13

i'm in the middle of something right now, but when i get a chance later today i will provide an example 1 frame mlv and receipt for mlvapp...

masc

Quote from: 70MM13 on January 28, 2019, 08:36:44 PM
i'm in the middle of something right now, but when i get a chance later today i will provide an example 1 frame mlv and receipt for mlvapp...
Thumbs up! Thx :)
5D3.113 | EOSM.202

70MM13

OK, I grabbed that same example from the resolve screenshot and loaded it up into mlvapp, and the result is very interesting...

It seems that mlvapp is more robust than resolve!  Really!!

Getting a good look on that shot in resolve (without using linear gamma) was quite difficult, and in mlvapp, it is comparatively easy.

Nevertheless, using the included receipt for linear gamma, you can see how nice it is to work with.

First, note the settings (disregarding colour temperature, tint, and the crazy curve).  Other than dropping the dark strength to 0, everything is default.  Does it get any easier?

Now, maximize the size of the edit window so you can make precise adjustments to the curve, and grab any point and make SMALL adjustments.  It's easy and intuitive to make changes to the image, and the curve almost tells you if you are going too far...

Keep in mind that this example is using the reduced gain, and the subject matter is not typical.  This is a really dark scene with extremely strong highlights.

Using this linear setup makes it really easy to work with a really difficult shot.

https://drive.google.com/open?id=1v4Jr-MT3YOrcLb3vtUp1Wz19OH9aAQ9r

masc

Thx! Nice to hear you like it. Interesting way of processing. I need to think about why there is indeed a little less noise, than using conventional method... I tried getting the same (or a very close) result using "Tonemapped" or "Film" - there is a very little more noise to find.
5D3.113 | EOSM.202

70MM13

I liked the results I got on that shot so much that I regraded the whole video using that receipt as the starting point, and it went very smoothly and easily, in fact, by far the best grading experience for me yet.

In a couple of shots with extremely bright highlights, I dropped the top of the curve by about 10% and added  a 45 degree bump at the top right side of the curve, and it made the highlights very nice looking!

Yes, the shadows look better this way, not just less noise, but also better "defined".

I really like this, and urge everyone to try it...

Maybe you can make a special "nonlinear" mode that takes advantage of this!

togg

For me the best results are starting with bmd film and have an output lut blackmagic cinema camera rec709 v2 :) It is good especially if your workflow doesn't roundtrip properly and you have to export prores444 masters.

But yea the resolve controls will react differently to different starting points, I didn't play much with linear but if it works why not :)

ilia3101

@70MM

Will take a look later, if it's really amazing(you mention defined shadows and nice highlights), I will try and recreate that curve as an MLV App profile.

masc

Quote from: Ilia3101 on January 30, 2019, 10:44:31 AM
@70MM

Will take a look later, if it's really amazing(you mention defined shadows and nice highlights), I will try and recreate that curve as an MLV App profile.
It is not just a kind of profile. It works only, if you set Dark Strength to 0 (because otherwise shadow data is cut before gradation curve can pull it). The next point is: I think it looks good, because you just have one tool only used now, which edits brightness values (->profile=off, 5 curves sliders=off). If this (as new) profile fits, it will be good as we see it here. If it does not fit, we will again edit using the curve tools and this will bring a little less quality.
So I think best should be to safe this receipt and using it as base for such processing tasks. And finetune it for special tasks.

I hope someone is able to understand what I wanted to write... haha.
5D3.113 | EOSM.202

70MM13

Yes, it makes sense; that's what I was guessing for the reason for the better results.

Curves always provide a better result because there isn't any "stairstepping" of the data.

I agree completely that it means the curve has to fit the data, seeing that I created the provided curve specifically to match the scene, and more importantly, the reduced gain iso "109" I was using.  It may not match stock iso.

But it would be easy to create a curve for (each?) stock iso.

The idea for me is to have a suitable curve as the starting point, and then provide a visually straight curve for the user to make adjustments afterward for corrections.  Of course, this will be an easy option when you add nodes.  At that point, the discussion is irrelevant.  The user can load the appropriate receipt, and then adjust a "fresh" curve in a second node...

I see a lot of people mentioning "filmic" looks where they are actually just trying to say "correct".  This is a good clear way to get that look, at least for me.

As a side note, I think that "sliders" would work beautifully if they always affected all of the data, thus avoiding the stairstepping problem, even if the effect if zero on some/most of the data.  No more halos etc.  Probably easier said than done, but this is why I avoid those kinds of adjustments as much as possible on every grading platform.

Resolve is simply terrible for halos.  Really bad.  Even the basic raw control for highlights creates very ugly halos.  Make the same adjustment manually by using a curve in another node... no halo.

masc

At least in MLVApp the "stairstepping" is identical between curve sliders, gradation curve and profile curve, because the realization behind the GUI is in some ways identical.

BUT: the less of these tools you use, the better the quality will be in the end. For example: profile=linear (means no math operation), Curve sliders strengths and lightening = 0 (means no operation), use just gradation curves ----> only one stairstepping effect --> good quality in terms of noise. The more of the tools you use actively, the worse the resulting quality will be (even if you might have to pixel-beep for it).
5D3.113 | EOSM.202

ilia3101

I bet this stair stepping quality decrease is not possible to notice with the naked eye

70MM13

It's very noticeable, in fact it is a big problem.

It's what causes the halos around bright objects in dark scenes when you pull down highlights with a slider.

Same thing with pulling up shadows, etc.

Very ugly, and very obvious.

masc

Quote from: Ilia3101 on January 30, 2019, 03:36:15 PM
I bet this stair stepping quality decrease is not possible to notice with the naked eye
Yes, I agree... mostly. I could imagine that in the very deep tones (where the log curve is steeply rising) it could make a tiny difference.

Quote from: 70MM13 on January 30, 2019, 05:24:09 PM
It's very noticeable, in fact it is a big problem.

It's what causes the halos around bright objects in dark scenes when you pull down highlights with a slider.

Same thing with pulling up shadows, etc.

Very ugly, and very obvious.
Now you are talking about something completely different. Shadows and Highlights has nothing to do with the topic we were talking above. There we were talking about curve sliders - this is Dark/Light Strength/Range and Lighten. The result of these 5 sliders is a kind of simplified gradation curve. When using them + Exposure you should get mostly the same results as with your posted linear strategy.
Shadows and Highlights should be used as last resort if nothing else helps. Somewhere else I wrote already: better try Exposure=-2EV & Lighten=50 as base setting.
5D3.113 | EOSM.202

70MM13

I see...

Whatever the cause may be, it's still best to do these things by hand with curves, at least for me.

Here is an example:
I didn't spend any time trying to match the exact results, I simply pulled the highlights down with the slider for the "slider" image, and then played with curves for a few seconds (literally) for the "curves" image.

Again, I was not trying to match the look, just show the effect...






masc

I tried getting as close as possible to your result without using linear profile. Some details are slightly different... but it is not far away I think.
https://www.dropbox.com/s/d198metcb17xgpt/nonlinear.marxml?dl=0

As I wrote: Shadows and Highlights is something completely different. It cannot be compaired at all.
5D3.113 | EOSM.202

70MM13

Nice job matching the look.

As you said, it's pretty close...

The shadows definitely lack the same richness, and there's more noise.  The noise shows a lot more in motion than with a single frame, of course.

There's definitely an advantage with linear, at least I think so!

ilia3101

Why can we not import clips in to a session twice?

masc

Quote from: Ilia3101 on January 30, 2019, 08:18:13 PM
Why can we not import clips in to a session twice?
Good question... this is more than a year ago I implemented that. If I remember right, I got issues when one clip was added twice, because the path+name is my unique identifier for a clip (matching between session list and receipts). For sure this can be implemented different, but this was the easiest and already hard enough for the beginning. But I am not 100% sure if this was the reason. We could just try to comment out the filename check and test when it crashes... :D

Edit: and another problem: batch export must be adapted, because a later imported clip overwrites the clip exported before. Okay... we could add something to filename. But this brings more problems than one might think... (will break e.g. logic of FCPXML assistants).
5D3.113 | EOSM.202

70MM13

I suppose you could work around it by copying the mlv and renaming it so you can load it again?

masc

Yes, doublicating and renaming always works best!
5D3.113 | EOSM.202

ilia3101

unfortunate that it is not easy :(

I guess renaming method will have to do