nature video using resolve ACES + REC2020

Started by 70MM13, November 02, 2019, 12:41:23 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

70MM13

This is by far the best setup I have ever tried, even better than the last one!

and it is unbelievably easy to use, while giving spectacular results...

under colour management:
input colour space : bypass
timeline colour space : ACEScct
output colour space : rec2020 HLG ARIB STD-B67
limit output gamut to output colour space
timeline to output tome mapping: luminance mapping

everything else default

yank the highlights down with both the camera raw control and the tiny input box at the bottom of the colour wheels page, and then push it hard.  it looks fantastic!

maybe a little better since it is iso 111 ;)

but don't tell the placebo salesmen.  they'll throw a fit.

https://www.youtube.com/watch?v=h_ch8UhDZWo

AF-OFF

Great Tommy! very useful info and awesome video!
May I ask which gear did You used?

Best
Afoff


Danne

Quote from: 70MM13 on November 02, 2019, 12:41:23 AM
but don't tell the placebo salesmen.  they'll throw a fit.
Personally I am only interested in true image refinements. My own tests reveals more than meets the eye and I have since left the iso_reg module for other work.
In your case. Placebo or not, do whatever you like but do not patronize other people's opinions.
On a sidenote. If you're interested in some deeper research around you're settings it could be done. I wouldn't mind to settle exactly what is going on imagewise with these regs. That is we can keep an objective mindset. A1ex provided all tools needed to measure any dynamic range refinements so shouldn't be too hard.


Luther

Oh yeah, now you're on to something. Great colors, I have no criticism to do, this is as good as it gets. On the 4m30s scene, there's clear color separation, even though the hue range between the red tree (on the right), the yellow tree (on the left) and your skin tones are very close, it still preserved the tones as they are. The shades of blue in your clothes are also well defined, without banding or posterization. This is really difficult to achieve if you don't have a good color manament system as ACES. Well done.
The contrast and sharpness is also spot on, not excessive or too flat. Some scenes have clipped highlights on leafs, but that is normal considering MKIII doesn't have a sensor as good as Alexa LF.

Quote from: Danne on November 02, 2019, 05:33:42 AM
Personally I am only interested in true image refinements. My own tests reveals more than meets the eye and I have since left the iso_reg module for other work.
In your case. Placebo or not, do whatever you like but do not patronize other people's opinions.
On a sidenote. If you're interested in some deeper research around you're settings it could be done. I wouldn't mind to settle exactly what is going on imagewise with these regs. That is we can keep an objective mindset. A1ex provided all tools needed to measure any dynamic range refinements so shouldn't be too hard.

Is it completely "placebo" though? I find it hard to believe all those graphs showing DR improvements are completely fake. A1ex himself said we only have 'rounded' ISO values (100, 200, 400, etc) because of industry standard (or measurements), not because they have the best SNR outcome. Maybe what @70MM13 is doing is not the 'correct' way, but I definetly think there's room for "true image refinements" in the ISO area, it just wasn't explored deep enough.

Danne

I am referring to my own tests. I didn't watch the movie and I am adressing the patronizing tone.
As I said. Some simple tests reveals what is what.

a1ex

Quote from: Luther on November 02, 2019, 06:06:54 AM
Is it completely "placebo" though?

This particular test was about 0.1 EV actual (measurable) improvement (from ISO 1600 to 1500), and about 3 EV of "placebo" or whatever you want to call it (because the actual ISO used to configure the sensor readout was different from what was printed on the screen - 1500 vs 100). My eyes are not good enough to notice a difference of 0.1 EV, no matter how hard I try to pixel peep ;)

On the other hand, ISO 111 reduced from Canon's ISO 200, or other low ISOs obtained from iso_regs.mo, are definitely not placebo. These do offer measurable improvements (see first post here), and I've never said otherwise. I am aware my measurement technique needs improvement, so my numbers might not be accurate, but the image quality improvement is definitely there.

Quote
but don't tell the placebo salesmen.  they'll throw a fit.

For the record, I've asked OP multiple times for some good test images - with specific requirements - in order to compare various ISO configurations, in the context of the placebo configuration linked above (the ISO 1500 mislabeled as ISO 100 in ML menu). The test scene that was provided by OP had an extremely tiny clipped area (the filament of a light bulb), where the highlights looked more or less the same at ISO 100, 400 or even 1600 (!), all other exposure variables being equal. That is, only a tiny percentage of the image area was overexposed, and that clipped area was not changing much, even if you increased the amount of overexposure by 4 stops (!), i.e. from ISO 100 to 1600. For that particular image, the best exposure decision was to use ISO 1600 or even higher, and at these higher ISOs, my tweaks do *not* provide any measurable improvement (at least not yet). ISO 100 would have still overexposed the filament, so using a higher ISO was a no-brainer there (again, in *that* particular test scene, for *that* particular ISO configuration - 1500).

Then, instead of providing good test images for a comparison, OP started to flood the forum with a bunch of crap statements like this one, effectively killing any motivation remnants I might have had to work on the ISO research topic. That's not how things are supposed to work here. Far from it.

You have been warned.

timbytheriver

@a1ex I'd be willing to set up a series of tests as per your request, so that we can finally sleep soundly in our iso-beds. It would be great if we could finally agree on the benefits of the research or else move on.  :) 

I've read your requirements but so that I don't waste a bunch of mine and others' time maybe we should all agree on the precise setup of the test scene?

Scene:

  • High-dynamic range, presumably? Be good to agree a fixed number of stops. Then I could measure this with my spot-meter and keep constant. Please suggest.
  • Approximate percentage of frame for highlights. I think @a1ex mentioned a much broader area then bare lightbulb. Please suggest.
  • Content? Objects, skintone? etc. Please suggest.


Method:

  • I find that the ML spotmeter and the raw histogram with OVER hints are not consistent with each other. What method should we use for accurate exposure?
  • Should the brightest area be set always at 100% or over/under? Use of dedicated meters for exposure. Spot/Incident?
  • Should shutter be fixed at a particular agreed setting? This will require use of NDs I presume to keep the exposure constant?
Resolution, Camera (I have 5D2 and 5D3), build version. ?

Anyone want to share suggestions or ideas?

5D3 1.1.3
5D2 2.1.2

Luther

Quote from: a1ex on November 02, 2019, 10:35:41 AM
On the other hand, ISO 111 reduced from Canon's ISO 200, or other low ISOs obtained from iso_regs.mo, are definitely not placebo. These do offer measurable improvements (see first post here), and I've never said otherwise.
Right, so the research was not a waste of time after all. There's room for real image improvements in the ISO research, that's good.
I'll keep myself out of this "war" now :P

Quote from: timbytheriver on November 02, 2019, 12:38:04 PM
Anyone want to share suggestions or ideas?
I don't have expertise on this, but I think you need a controlled environment, like a closed box with a lamp inside, so that luminance keeps constant. I would also suggest to keep aperture constant, because some lenses don't have accurate aperture rings (particularly old manual lenses). Oh, and I think there was something about manual lenses or too bright lenses (>f/1.0?) that Canon automatically makes the image drop white point (or add +1 digital gain, idk) or something, @a1ex can correct me if I'm wrong.

timbytheriver

@Luther

Thanks! Good catch about lenses. So many variables...  :-\
5D3 1.1.3
5D2 2.1.2

Danne

There's the module raw_diag.mo from a1ex to check dynamic range(thanks for clarifying iso 111 @a1ex).
Hopefully I can take a look again in the future. Good luck @timbytheriver. Hopefully you'll be able to dig out some more here. Seems regs are there in both adtg_gui.mo and iso_reg.mo so it should be a matter of time and patient testing going forward.

timbytheriver

Quote from: Danne on November 02, 2019, 02:25:19 PM
Hopefully I can take a look again in the future. Good luck @timbytheriver....

@Danne Thanks! The future is bright; about 0.35 EV brighter than the present hopefully...  :P

This may now be the wrong thread in which to continue this discussion – I'll let the mods decide.

It appears from @a1ex's last post that lowering iso from a base 200 to 111 via the already discovered cmos and adtgc registers is now not disputed as having a positive effect on the image. That means that it is pointless testing this combination any further, yes? Are we now at a point where the only testing left to do is on much higher ISOs to see whether a similar improvement can be achieved?

I've been through the entire iso gain thread https://www.magiclantern.fm/forum/index.php?topic=10111.0 again today, and apart from my head spinning, I feel that there maybe needs to be a 'state of the nation' summary post that outlines clearly:

1) Where the research is at, as of today.
2) What the agreed conclusions are.
3) Defining a new goal for the research.
4) What remains to be done in terms of testing.

Then maybe the thread could be locked/frozen until the Phase II of the research data starts coming in.

It seems that this is such an important gain (fnah!) for ML that it would be a shame to see it lost in a civil war between the INTJs and the ISTPs of this community. We all have something to bring to the table.


5D3 1.1.3
5D2 2.1.2

DeafEyeJedi

Quote from: Danne on November 02, 2019, 02:25:19 PM
There's the module raw_diag.mo from a1ex to check dynamic range(thanks for clarifying iso 111 @a1ex).

Ah, just like the good old days... I'm happy to refresh my memories on this module along w you @timbytheriver!

Quote from: timbytheriver on November 02, 2019, 04:22:45 PM
It seems that this is such an important gain (fnah!) for ML that it would be a shame to see it lost in a civil war between the INTJs and the ISTPs of this community. We all have something to bring to the table.

Agreed.
5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109

ngemu

the colors are so nice, amazing work! What lens are you using?

70MM13

thanks!
most of it was a rokinon cine 35mm T1.5, but the closeup of the buds was a canon 50mm f/1.4

70MM13

Quote from: AF-OFF on November 02, 2019, 02:51:10 AM
Great Tommy! very useful info and awesome video!
May I ask which gear did You used?

Best
Afoff

thanks!

it's a 5d3, and an umbrella ;)

Aperture Science

The color science and the quality of the image is awesome !
Glad to see these wonders :D

70MM13

i couldn't agree more about these wonders!

it's unbelievable what we can do today.  it is too easy to lose sight of the incredible capabilities we have with these toys of ours :)

i love it.