Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Curious Case of Benjamin Buttons

Emmanuel Cambier said:
First, thanks to OFFHOLLYWOOD for providing the wonderfull image I have been posting earlier, and to DIGITAL FX for hosting it.

Well I am ready to face the fact that the RedOne is not without limit.
In all the pictures shot on the camera, I have laid my greedy hands on, I have noticed that there is 2 kinds on cliping.
The obvious one is in the highlights of course, and the obvious fix is to only allow white regions presenting no noticable details, to clip.
The less obvious cliping that red shows is in the black zones of an image. It is true that some zones don't have any details in it, just a noisy grey, with no cure in sight. Come on guys check all the footage you want, and you will find those zones. So I would agree with Claudio that there is a limit to the amount of informations in dark unlit areas, and when you reach it it's like cliping your highlights, no way back, i'm facing the truth here.

Where I would tend to disagree with Claudio is in the conclusions one could draw from such discovery.
It seems that this limit hapens when the sensor will output a luminance of around 25 (RGB). 25 seems to be the lowest luminance reading in all uncorrected images downloaded, 255 being the highest.
If you leave those dark zones as is you have no detail and a lot of noise, and if you lower it by approx. 10 you are left with zero noise and still no detail, but your zone now looks like pure black (still reading 15 in luminance).
So, to me, the problem is where in the picture are located these zones in terms of exposure.
From all the images I've seen I would tend to conclude that these zones happen to fall where a deep black is expected if not desired.
Therefore I wouldn't object them as long as the Display system can reproduce blacks in a satisfying manner, otherwise you're stuck with grey patches in your images.

Now maybe there is something for Red to look for in those blacks, we shall see.

Sincerly yours

Emmanuel
Click to expand...

Emmanuel, this is the kind of post I appreciate; very helpful indeed. Merci!
 
That's how all sensors work. They clip at the highlight end, and just fade into noise as you get darker and darker. The limited between where you can discern detail in the shadows and detail in the highlights is the signal to noise ratio or dynamic range.

The actual level at which this occurs is not really important as long as we're giving you the full precision from the sensor to a high bit format. Noise has a distribution, both positive above black, and negative below black. Yes, we preserve and give you this below black, so as to preserve the entire noise floor of the sensor. If you try to look below this nominal level of black, it's pure noise and no detail at all, but clipped noise can look unnatural so we preserve it, and then you can either clip it out, or crush it on a curve. Your choice.

Graeme
 
isn't 0 -255 8bit scale ?
... these are 12bit files, so should we be talkin 0-4000 scale ?
 
Robert Sanders said:
Even if the Viper had more shadow detail, in my opinion it doesn't matter, as those areas are fraught with noise that is really unpleasant looking. Again, my opinion, but I really do not like the noise characteristic of the Viper's noise floor.
Click to expand...

But Claudio is obviously of the opinion that the RED One doesn't have enough dynamic to perform with as much "malleability" as the Viper. Here's his quote:
"The files I get from the Viper/F23 look pretty flat so the image is very malleable. The more dynamic range I have, the more chances I have of bringing down a blown out sky or lifting something out of the blacks".

Could someone please expand on the phrase "lifting something out of the black' as it relates to the RED ONe?
 
Graeme Nattress said:
That's how all sensors work. They clip at the highlight end, and just fade into noise as you get darker and darker. The limited between where you can discern detail in the shadows and detail in the highlights is the signal to noise ratio or dynamic range.

The actual level at which this occurs is not really important as long as we're giving you the full precision from the sensor to a high bit format. Noise has a distribution, both positive above black, and negative below black. Yes, we preserve and give you this below black, so as to preserve the entire noise floor of the sensor. If you try to look below this nominal level of black, it's pure noise and no detail at all, but clipped noise can look unnatural so we preserve it, and then you can either clip it out, or crush it on a curve. Your choice.

Graeme
Click to expand...

Hi Graeme

How about a tool in RedCine that would allow us to bring those blacks (and the noise) down were they belong without altering the rest of the image's brightness.
I know curves could do that, but I'm thinking of something more straight forward (pun intended).

In Aperture I use the brightness control exclusively for that purpose, it's forking fine since it leaves the highlights intact (wich i boost with exposure 's slide), but the medium part of the image can end up a bit under exposed.

I don't have access to RedCine or Red Alert cause I'm on a G5, so I have to use my imagination, maybe it's all in ther allready.

For Everybody, I'm more than happy about red footage and noise level and shadow details and all…
But Claudio is not, so I think it's worth trying to understand his point of view wish I don't share, and try to find out if there is something missing on the bottom end of our beloved cameras.
I don't think therse is anything missing, but let's all check it, can't hurt.

Yours

Emmanuel
 
Claudio hasn't shot extensive tests of the current RED ONE yet so I would stop beating up on the guy for not having a definitive opinion. He may change his mind on the next digital feature, who knows. You guys are getting way to sensitive about this.

Don't forget that all of "Zodiak" went through an expensive but advanced noise reduction process at John Lowry's company, normally used for film restoration work. With that tool in your arsenal, you may be more comfortable with detail in the shadows that has some noise in it.

Clearly Claudio is more comfortable working with 10-bit LOG images and hasn't yet explored the issues regarding working with the RAW files from the RED camera -- he probably saw some RGB conversions from RED ALERT that looked a bit limited in terms of latitude, as we have all seen in some people's test footage. I'm sure on some future project, he'll take the time to explore it further. I'd stop with the nitpicking of everything he is saying based on an initial reaction.
 
By definition, a blown sky has blown and there is no more detail there. It doesn't matter how many stops DR your camera has - if you blow the sky out - ie, it clips, there is not data there to pull back. None. Not on any camera.

Video cameras that shoot "raw", like the Viper use a log curve to fit all their sensor dynamic range into a 10bit range to go over HD-SDI. That, if it's done so as to preserve all the DR, will look very flat and boring (and green too, if it's the Viper).

RED is different. We don't shoot to legacy video tape, over legacy serial digital video interfaces. We shoot data and store data. The data is stored RAW, and then "Developed" in post. That developing process can keep all the DR the sensor recorded or you can throw some away or make the image look all pretty, or whatever you feel like.

In REDAlert / REDCine, you can usually get some detail back in "blown" highlights as they might not actually be blown. The default "development" has some "headroom" in there to help you keep a little extra highlight detail, but it's there if you set the exposure in our apps lower.

Lifting out of the blacks is just putting a secondary colour correction or a primary colour correction curve to boost the brightness of dark areas of the image to make them more visible. By default, the log curve we use, and the way we use the REC709 gamma curve show you ALL the dynamic range, and blacks are quite boosted by both those curves rendering all their detail reasonably visible to begin with.

That's why we don't generally measure of analyze dynamic range with nice images, but with calibrated transmissive charts so we can really see what's going on. You can read about that on other postings by me though.

Graeme
 
Emmanuel, exposure always crushes / clips highlights, so if you want to see / rescue what's on in the top end, boost brightness, reduce exposure. To nicely blacken up the shadows, increase contrast, or, for finer control, play with curves.

To further David's point, looking at a developed image in REDCine/Alert can be like looking at a positive print made from the negative, or it could be like looking at a scan of the negative, or somewhere inbetween. Obviously, if you judge the DR of film by looking at a print you'll think it has a much lower DR than if you look at the scan of a negative.

Graeme
 
isn't 0 -255 8bit scale ?
... these are 12bit files, so should we be talkin 0-4000 scale ?
Click to expand...

Besides the range difference, you have non-linear vs. linear information. When working in 8-bit, you are working with non-linear values. The 12-bit from the sensor is linear.

One of the most exciting aspects of the RED camera, to me, is the fact that we are getting access to the unadulterated sensor data. This is, of course, if you don't count compression.

Claudio keeps pointing to compression as a possible cause. One way to dispell this is to actually provide a frame of truly uncompressed 4k along with a compressed version.
 
David Mullen ASC said:
Don't forget that all of "Zodiak" went through an expensive but advanced noise reduction process at John Lowry's company, normally used for film restoration work. With that tool in your arsenal, you may be more comfortable with detail in the shadows that has some noise in it.
Click to expand...

True. I forgot to mention that. It's a noisy camera.
 
Concrete said:
I don't get it, Emmanuel. What did you expect? Infinite shadow detail?
Click to expand...

If I'm understanding correctly he is saying that values below 25 (this is if you take a file into photoshop and use the eye-dropper tool across 256 possible values for example, in 12 bit land that would be 400 aye.) Thats 10% of the total signal... anything below that is mud. Even if it's not in the noise floor. NOT TRUE friends. Emmanuel look closely at that histogram, that tiff has been graded.

Lets start fresh.

Here's a shot I did, at night, all existing light. Wide open on and 18mm Superspeed. 355/360 shutter.

I took the .R3D into RedAlert and exported it with these settings.

ISO 320
5000K
Tint 0
REC 709
Exposure 0
Saturation 1
Contrast 0
Brightness 0
RGB all 1
DRX 0.50
Black offset 720
Flat curve.

The tiff file is here: http://www.redhax.net/NightShot.tiff (Please do mirror ;)
http://www.bealecorner.org/red/RED-NightShot.tiff

And here's a JPEG:
582_1196107578.jpg

and the histogram:
582_1196107966.jpg


Take it the tiff into photoshop. Open levels and set the RGB levels to 0,1.0,25. Here is what you get. I used the color range selection to show whats above 25.
582_1196107233.jpg

Maybe there is a flaw in my methodology here but I hope that prooves to you there are details in the blacks on RED below 10%. Tons.

And keep in mind this is Redcode compression part of 4 minutes of material recorded to an 8GB flashcard.

IBloom
 
David Mullen ASC said:
Claudio hasn't shot extensive tests of the current RED ONE yet so I would stop beating up on the guy for not having a definitive opinion. He may change his mind on the next digital feature, who knows...
Click to expand...

Good point. And let's not forget that all RED ONE cameras are being updated with a new daughter board to the sensor. This had been stated by RED to result in "increased dynamic range and reduced noise" and be a "major breakthrough".

Original thread:
http://www.reduser.net/forum/showthread.php?t=5454

I have a feeling that this may be the "tweak" that helps find the Cameras sweet spot and those really concerned with DR should wait to see images from these updated cameras. Hopefully these cameras will be out in the field soon.

Anyone who has one yet?
 
You might even get a tad more in the highlights if you set the exposure a little lower in REDAlert. You can take it down to -0.4 or more depending on the white balance.

Graeme
 
They tested Red, Dalsa, Viper, and the D20, so it follows that he isn't unaware of raw imaging etc... look forward to your results (when using the newest version cameras).

ps - Is Brad Pitt as handsome as they say ? :biggrin:

;)
 
Gbabymogul said:
They tested Red, Dalsa, Viper, and the D20, so it follows that he isn't unaware of raw imaging etc... look forward to your results (when using the newest version cameras).
Click to expand...

They did not test RED One before CCBB.
 
Linear Light, REDLog and REC709 gamma all preserve the dynamic range. Use whichever is appropriate for your output.

Graeme
 
You must log in or register to reply here.
Back
Top