Monstro: High ISO grid pattern

[Chris McKechnie]

Any chance you did this test with Dragon (and even MX) as well? This is a can of worms that seems like a bit of a nightmare to have to test... Especially since the sensor temp isn't what is controlled by the temp control under fan settings.



Great test, but also a giant pain in the ass... Like it could be different per camera, and those results vary quite substantially for such a small shift in temp.

Yup..which is crazy. It shouldn't be fluctuating that much. I think RED needs to come out with a statement stating some facts about noise. There are just too many inconsistency's across the board which makes people frustrated. The vertical noise pattern is an issue that has plagued me for a long time with no end in sight. Even with manual black shades, exposing properly, etc., with the latest sensor tech, it just never got right. MX and dragon sensor were far more stable in that regard.

 

[Les Hillis]

Then after you pick your preferred temp, do four consecutive blackshades at that same temp and pick the best one from those. :biggrin:

 

[Björn Benckert]

Not so great test if you ask me. The thinking behind it is wrong.

You should not be too sure 39F and 41F looks "better" it very likely so that the calibration possibly shows them a hint darker.

But "darker" is not always "better". Darker is darker thats all, if you even them all in terms of gain, then they all look pretty much the same.

And if you think of it thats actually what you need to compare. As an image graded pitch black with no noise is simply not better than the same image graded a bit bright with noise... And thats all that comparison actually show.

Here is an attempt to even out the brightness between them just in PSD with the JPG, then the 39, and 41 gets pretty much the same noise pattern.

LEVELED OUT by Björn Benckert, on Flickr

 

[Mike P.]

Thats wrong thinking.

Dont be too sure 39F and 41F looks "better". just as the calibration possibly shows them a hint darker.

Darker is not always better. Darker is darker thats all, if you even them all in terms of gain, then they all look pretty much the same.

And if you think of it thats actually what you need to compare. As an image graded pitch black with no noise is simply not better than the same image graded a bit bright with noise... And thats all that comparison actually show.

Here is an attempt to even out the brightness between them just in PSD with the JPG, then the 39, and 41 gets pretty much the same noise pattern.

LEVELED OUT by Björn Benckert, on Flickr

Whew, touche. Was definitely a little knee-jerk reacting there (I hadn't downloaded the files yet).

Though I am curious what variable would've made the "cleaner" temps blackshade darker (which, to be honest, I would prefer as it'd be perceptively cleaner on the day/videovillage)...

To the OP: just checked the .r3d. No grid in RCXp a Full/100%, so as Graeme mentioned, it's likely a downscale behaviour in Resolve.

 

[Björn Benckert]

Whew, touche. Was definitely a little knee-jerk reacting there (I hadn't downloaded the files yet).

Though I am curious what variable would've made the "cleaner" temps blackshade darker (which, to be honest, I would prefer as it'd be perceptively cleaner on the day/videovillage)...

Its not cleaner its basically same image developed a hint darker or brighter, and its by a small fraction, I doubt the clients in video village will see the difference. I guess this is looking at black shades at 12800 iso or such.... thats not optimal iso to impress video village, if you slide ISO at least down to half of that then your blacks will probably show as proper black. But preferably give your lowlights a bit more fill and push down with a curve to compensate and your blacks should not even a trace of noise.

 

[Patrick Tresch]

Great test, but also a giant pain in the ass... Like it could be different per camera, and those results vary quite substantially for such a small shift in temp.

This test would proove interesting if you get twice the same results with a same temperature setting. I bet it's less scientific as we would love to see.

 

[luigivaltulini]

guys, are you talking about this grid?
this appears many times if I increase iso and the condition of light is very low and I apply a slight denoise.Without denoise and less perceptible.
I will never get to these ISOs, I just wanted to understand what the hell it was, and I never investigated it.

Screen Shot 2019-04-02 at 23.13.17 by luigi valtulini, su Flickr

 

[Mike P.]

guys, are you talking about this grid?
this appears many times if I increase iso and the condition of light is very low and I apply a slight denoise.Without denoise and less perceptible.
I will never get to these ISOs, I just wanted to understand what the hell it was, and I never investigated it.

Screen Shot 2019-04-02 at 23.13.17 by luigi valtulini, su Flickr

Yeah, that's looks like the same kind of "grid" we're discussing...

It has to do with the downscale method (and/or potentially viewing software... or monitor resolution not being 8k native), and the .r3d footage doesn't actually have it (it's typically not the sensor).

This test would proove interesting if you get twice the same results with a same temperature setting. I bet it's less scientific as we would love to see.

Yeah, Bjorn cleared it up with his pic... Actually, gonna edit my post so its less "sky is falling" in tone.

 

[Blair S. Paulsen]

There's plenty of folks making incredible images who have no idea how a downscaling filter functions. That said, when authoring an image, I regularly audition several downscaling options to see how I might "play" it to get the look and feel I'm after. TBH, I started out auditioning downscaling filters in search of the ideal balance between sharp and smooth. Mostly pixel peeping comparisons for edge transitions, but over time I started noticing more global influences.

This experimentation also led me to a deeper appreciation for how much one's perceptions are dependent on the totality of the monitoring environment. How large is the screen? How close am I to it? Is the room dark? Resolution? Emissive or reflected? OLED or LCD? Color bit depth? Peak white level? Etc.

Cheers - #19

 

[Björn Benckert]

There's plenty of folks making incredible images who have no idea how a downscaling filter functions. That said, when authoring an image, I regularly audition several downscaling options to see how I might "play" it to get the look and feel I'm after. TBH, I started out auditioning downscaling filters in search of the ideal balance between sharp and smooth. Mostly pixel peeping comparisons for edge transitions, but over time I started noticing more global influences.

This experimentation also led me to a deeper appreciation for how much one's perceptions are dependent on the totality of the monitoring environment. How large is the screen? How close am I to it? Is the room dark? Resolution? Emissive or reflected? OLED or LCD? Color bit depth? Peak white level? Etc.

Cheers - #19

Yes agree, I see lots of footage now coming from some of the most respected colorists out there. And I started to note a pattern. My theory is this. They still sit with their dolby 2k screens and grade, some probably do the grade at sub native camera input resolution / proxies view and then only when exporting they turn on full debayer full quality and roll it out in UHD or what ever is asked for. Well, when I get the stuff and look at it in a straight UHD pipe to a large screen I can not simply believe that they actually wanted to add and except the amount of noise they let trough, as some of them simply now go way over what they done in the past when they where delivering 2k files.

So my theory is that as they view the stuff in less then their delivery format they tend to add and except more noise as its simply showing less on their screens. So sure is a big need for large at least UHD grading screens.

 

[Tim Sessler]

Its not cleaner its basically same image developed a hint darker or brighter, and its by a small fraction, I doubt the clients in video village will see the difference. I guess this is looking at black shades at 12800 iso or such.... thats not optimal iso to impress video village, if you slide ISO at least down to half of that then your blacks will probably show as proper black. But preferably give your lowlights a bit more fill and push down with a curve to compensate and your blacks should not even a trace of noise.

Interesting! Obviously this wasn't meant to show how ISO 12,800 looks but rather to push the exposure to the max (12.8K ISO +8 Flut) to see if there is any difference between different sensor temps (as was indicated by other users in past threads).
I'll happily rerun the tests and see how things change.


Why would you switch scaling to smoother? As per the Resolve Manual - which is not very specific or detailed unfortunately:

Sharper: Usually provides the best quality in projects using clips that must be scaled up to fill larger frame size or scaled down to HD resolutions.

Smoother: May provide higher quality for projects using clips that must be scaled down to fit a SD resolution frame size.

Also curious how the different LUT pipeline would affect the scaling or image quality?

And coming back to why I started this thread initially: how can I get the most out of this beautiful sensor. Even though not practical or needed for most projects, I enjoy rating the sensor much higher, underexposing it and having a lot of "natural" grain in the image instead of adding a post grain. I was just testing to see how a daylight exterior rated at 6,400 would look. Regardless it seems like the grainy image revealed a bigger finishing/downscaling issue. There is no need to finish in 8K right now. 4K is more than enough for what I am using the camera for. If Resolves scaling is the actual culprit - what better options are out there? I could easily finish a 8K Master in Resolve and downscale with another tool that is more sophisticated.

Curious to hear what your guys's workflow is!!

 

[Tim Sessler]

Also meant to share some stills from the high ISO test that I did. Just a quick walk in Santa Monica. All rated at ISO 6,400 - was using a Kodak Portra LUT, otherwise straight out the camera. Here are some 8K JPEGs: https://www.dropbox.com/sh/h8lf5996nvkazav/AAAj23Dk5b1QwwVWh5VCa6_Ya?dl=0

 

[Brendan H. Banks]

These look really great! Mind if I ask what Portra LUT you're using? It is for IPP2 or to use on top of Rec709?

 

[Mike P.]

I gotta ask; the sun was out so how come your cranked it to 6400 (or were you specifically testing high ISO)?

(Definitely love that under the boardwalk/sun peaking through shot. So spicy.)

 

[Tim Sessler]

These look really great! Mind if I ask what Portra LUT you're using? It is for IPP2 or to use on top of Rec709?

Thank you! It . works with REDWideGamutRGB and Log3G10. Would be nice to have it work within the IPP2 pipeline as well - thats my next task

 

[Tim Sessler]

I gotta ask; the sun was out so how come your cranked it to 6400 (or were you specifically testing high ISO)?

(Definitely love that under the boardwalk/sun peaking through shot. So spicy.)

Ha! Good point! I just finally got my ND 3.0 and still have to add more to it

The whole point of the test was to see how the image holds up at higher ISOs. I am always adding grain for everything in post - so I was curious if I could get a similar result by underexposing a ton and pushing the image to a point, right before it falls apart. I obviously understand that post grain gives me a lot more flexibility and is a much safer choice - but wanted to test it regardless