skintone

[paulherrin]

everybody chill... jake, have you read through the thread? there's actually some decent and balanced conversation and i'd much rather listen to you actually say something here.

 

[Mike P.]

But it's not less or more valid than putting mid grey at 4.5% in the raw, and having two more stops above mid grey (and correspondingly two more stops above 100%) with an ISO of 800. Of course, ISO is not determined by raw code values at all, but by the code values in the developed image, mapping 18% light to ~44% in the image.

Graeme

But doesn't it make noise more apparent in what will be mid values (which make up the majority and arguably the most important aspects of the scene) by mapping mids into the lower, more truncated bits of the capture (where as the highest stops will have 2048, 4096...)

So does that mean, under linear gamma space 18% light = 18% grey at around the film equivalent of 200asa? Also, isn't Linear gamma space in RCX essentially mapping the light values to their equivalent developed image percentages? E.g. 100% light = 100% image, 44% light = 44% image. And if so, why wouldn't that be a viewable option in camera (what's the disadvantage to the end user)? If you could learn to read that properly it'd be bullet proof, and there'd be no need to expose at ISO800 in order to retain highlights since all ~13.5 stops would be viewable all the time.

Speaking of which, would it not make more sense to have more info captured in the dark sections of the image than in the highs (which generally don't have important info)? Is it not possible to actually capture a negative image digitally, where the higher bit values (with more shades/data available) are reserved for the dark portions of the image? That way, you could actually pull more out of the shadows, which generally have useful subject/image material than highlights. The highest few stops are roll-off and don't necessarily need detail... but the bottom few stops could potentially have something worth seeing (aka something worth pulling up), and even if they don't, the extra info would be useful to squelch noise.

 

[Scott Crawley]

You're correct about pointless argument. It's pointless trying to have an unbiased conversation on this site...

Jake, we're RedUsers trying to make the best pictures possible with our tools. I know you have insights that could help us do just that, so why not share those instead of arguing about which is better? Hell Scott seems to agree with you so I'm not real clear what the problem is.

 

[Zakaree Sandberg]

Lov'n this thread.. thanks guys

 

[Jeremy Hildebrant]

I have a problem with skin tone being too magenta under tungsten light (Mole&Arri fixtures) even when white balancing off of a macbeth card. The white balance inevitably end up shifting the tint towards green and kelvin towards 3000ish when neutralizing whites/greys.. But nothing I can do as a primary adjustment in RCX gets the magenta out of the skin with out throwing everything elso off. Interestingly though I can bring it into Premiere Pro and do a simple Hue angle rotation about a negative 6 degrees and everything pops into place. Is there a way to make a similar hue adjustment with the just the controls in RCX? I assumed the Tint slider in combo with the color wheels in RCX would do it, but hasn't for me.

On a side note, when I throw an 80C in front of the lens and balance for that the magenta issue in skin seems to go away. Why?

 

[Nick Shaw]

In the camera's newest firmware Sony added a "High Key / Low Key" toggle...

...I would love to see RED incorporate a High / Low raw view mode in an upcoming firmware release.

The Sony High Key / Low Key toggle is essentially cycling the ISO development curve down a couple of stops then up a couple of stops. You could do a similar thing on a RED by rolling the ISO back and forth.

 

[Noel Sterrett]

On a side note, when I throw an 80C in front of the lens and balance for that the magenta issue in skin seems to go away. Why?

The 80C shifts the tungsten scene closer (but not all the way) to the fixed daylight balance inherent in the sensor CFA.

Cheers.

 

[Donald Strubler]

You're correct about pointless argument. It's pointless trying to have an unbiased conversation on this site...

+10000

 

[Lliam Worthington]

But doesn't it make noise more apparent in what will be mid values (which make up the majority and arguably the most important aspects of the scene) by mapping mids into the lower, more truncated bits of the capture (where as the highest stops will have 2048, 4096...)

Speaking of which, would it not make more sense to have more info captured in the dark sections of the image than in the highs (which generally don't have important info)? Is it not possible to actually capture a negative image digitally, where the higher bit values (with more shades/data available) are reserved for the dark portions of the image? That way, you could actually pull more out of the shadows, which generally have useful subject/image material than highlights. The highest few stops are roll-off and don't necessarily need detail... but the bottom few stops could potentially have something worth seeing (aka something worth pulling up), and even if they don't, the extra info would be useful to squelch noise.

Something to consider may be that as we see considerably more gradations in the higher end of the spectrum than in the lower, we are perhaps more sensitive to poor highlight falloff than better shadow detail?

 

[Gunleik Groven]

Something to consider may be that as we see considerably more gradations in the higher end of the spectrum than in the lower, we are perhaps more sensitive to poor highlight falloff than better shadow detail?

Fun part is that as opposed to sensors, we don't really...

 

[Graeme Nattress]

But doesn't it make noise more apparent in what will be mid values (which make up the majority and arguably the most important aspects of the scene) by mapping mids into the lower, more truncated bits of the capture (where as the highest stops will have 2048, 4096...)

So does that mean, under linear gamma space 18% light = 18% grey at around the film equivalent of 200asa? Also, isn't Linear gamma space in RCX essentially mapping the light values to their equivalent developed image percentages? E.g. 100% light = 100% image, 44% light = 44% image. And if so, why wouldn't that be a viewable option in camera (what's the disadvantage to the end user)? If you could learn to read that properly it'd be bullet proof, and there'd be no need to expose at ISO800 in order to retain highlights since all ~13.5 stops would be viewable all the time.

Speaking of which, would it not make more sense to have more info captured in the dark sections of the image than in the highs (which generally don't have important info)? Is it not possible to actually capture a negative image digitally, where the higher bit values (with more shades/data available) are reserved for the dark portions of the image? That way, you could actually pull more out of the shadows, which generally have useful subject/image material than highlights. The highest few stops are roll-off and don't necessarily need detail... but the bottom few stops could potentially have something worth seeing (aka something worth pulling up), and even if they don't, the extra info would be useful to squelch noise.

Not really on the noise. Indeed that's why we set recommendation at 800 for M-X just because there's low enough noise to more than justify it. Linear gamma is indeed a straight mapping, but try it - it's absurdly dark and hence utterly useless for in-camera monitoring. Without a well defined curve you just can't see all the stops a good camera can produce, and even then it's very very hard to show lots of stops with decent contrast, hence the use of log encodings to see everything, but at the expense of decent contrast.

With regards to capturing more or less info in highlights v shadows - you don't get a choice because sensors and AtoD's (generally) don't work that way. Indeed talking about bit levels is utterly irrelevant to the final image because that is defined by noise, not bit levels. It's common enough to gain up an image to see noise, but I've never seen anyone do extreme levels on an image to reveal quantization error (not having enough bit levels), other than when in an early development there was an error and redcine went through a 10bit linear representation and that did indeed induce visible quantization error. In other words, there's no benefit to bit depth beyond the noise floor.

Graeme

 

[George A.]

The 80C shifts the tungsten scene closer (but not all the way) to the fixed daylight balance inherent in the sensor CFA.

Cheers.

This helps A LOT.

 

[Björn Benckert]

I think to improve the view of exposure it would be good to have:
1. A high / low key toggle. Where gamma is adjusted up/down, since sliding the iso is not the same since iso is a linear gain and not showing the DR in the same way as a gamma correction.

2. A RGB toggle view where we can see the red, blue and green channels sepperatly as black/white hicons. Normally bad looking tungsten footage are usually due to a starved blue channel. Not having any exposure in blue gives 2/3 of the color range which leads to flat / smeared skin tones.

If / when the B channel is starved there is a way to repair it in post, basically the math used is to replace the B channel with the G channel and multiply the difference of G and B. This is is something that Graeme maybe could be implement as a "low blue rescue" function in RCXP. It's offcourse not an optimal approach to shoot in such way that the B channel is starved but to me that is the most common factor for footage thats difficult to dial in the skintones for.

 

[Jeremy Hildebrant]

This helps A LOT.

I understand why the 80C helps with noise, if there is enough light not to starve all the channels with 1 stop loss you get from the 80C, but what is the technical explanation for the hue shift towards magenta in skin with out it?

 

[Mike P.]

With regards to capturing more or less info in highlights v shadows - you don't get a choice because sensors and AtoD's (generally) don't work that way. Indeed talking about bit levels is utterly irrelevant to the final image because that is defined by noise, not bit levels. It's common enough to gain up an image to see noise, but I've never seen anyone do extreme levels on an image to reveal quantization error (not having enough bit levels), other than when in an early development there was an error and redcine went through a 10bit linear representation and that did indeed induce visible quantization error. In other words, there's no benefit to bit depth beyond the noise floor.

Graeme

Well then why would we expose to the right (put as much image into the higher/bigger bit stops) if the amount of bit levels plays such an unimportant roll? And, yes, I concur that linear is too dark to view, but what about if a "linear gain" was applied to the monitor path just to make it visible... not so much a gamma/curve, but just brighten the image so it's viewable (or is that exactly what the "raw view" is)?

Not to change the subject too much, but could you also explain the differences between REDgamma (1.0) and RG2/3 in terms of curve applied? I know RG3 is supposed to be more like Rec709 and represent contrast more accurately(?), but I consistently find REDgamma1 gives a softer highlight roll-off (at the cost of clipping a little earlier?) and that those highlights (and even the hadows) retain more saturation than RG2 and 3... Honestly, I was kind of surprised. Similarly, RC1 seemed to do "better" (warmer) in the skin tones (at the cost of colour accuracy.

 

[Graeme Nattress]

Well then why would we expose to the right (put as much image into the higher/bigger bit stops) if the amount of bit levels plays such an unimportant roll? And, yes, I concur that linear is too dark to view, but what about if a "linear gain" was applied to the monitor path just to make it visible... not so much a gamma/curve, but just brighten the image so it's viewable (or is that exactly what the "raw view" is)?

Not to change the subject too much, but could you also explain the differences between REDgamma (1.0) and RG2/3 in terms of curve applied? I know RG3 is supposed to be more like Rec709 and represent contrast more accurately(?), but I consistently find REDgamma1 gives a softer highlight roll-off (at the cost of clipping a little earlier?) and that those highlights (and even the hadows) retain more saturation than RG2 and 3... Honestly, I was kind of surprised. Similarly, RC1 seemed to do "better" (warmer) in the skin tones (at the cost of colour accuracy.

ETTR gives benefit of lower noise, not by devoting more code values to the signal.

Linear gain won't work, because then you loose the highlights through clipping. Of course, you could "bend" the gain, and if you bend it smoothly enough to avoid artifacts, lo-and-behold, the shape of the curve is now a gamma curve Yes, the RAW view is basically that - just enough to view the linear image without looking absurdly dark and without crushing the highlights.

RG1 was based on original REDLog. RG2 is similar, but based on the REDLogFilm starting point. RG3 is also log based, but adjusts the shadows more towards a fully developed look. Yes RG1 has more contrast in the highlights which could be perceived as greater saturation. Of course, it therefore shows less detail in the highlights.

Graeme

 

[paulherrin]

I understand why the 80C helps with noise, if there is enough light not to starve all the channels with 1 stop loss you get from the 80C, but what is the technical explanation for the hue shift towards magenta in skin with out it?

 

[Rudi Herbert]

So we are then, forced to ETTR if we want the maximum of information in our footage? In the R1, with is RAW exposure meter, a feature I miss on Epic btw, I expose just so that my bars reach the yellow or orange bars, right before red which is when channels begin to clip. That is the same as ETTR, correct? On land, this works great, but in situations/environments where there is a BIG unbalance among the channels, such as underwater for example, with very little red and lots of blue, ETTR yields weird hues and unnatural casts. Then again, exposing at less than that yields more natural colors but very little detail in midtones and almost none if the shadows and we end up with very dark images. Maybe having a gamma for such situations, with very low contrast and/or abnormal channel unbalances might help mitigate such problems, including skin tones shot with very little or no light against a dominant background that is not in the same spectrum as the skin?

 

[Rudi Herbert]

I understand why the 80C helps with noise, if there is enough light not to starve all the channels with 1 stop loss you get from the 80C, but what is the technical explanation for the hue shift towards magenta in skin with out it?

Exactly, we get s similar phenomenon underwater, with blue water heavily tinted towards magenta, and then, if you add a color filter, which is very close to an 80C albeit a bit darker, things get actually worse unless you are in very shallow water where the color balance is a bit more normal...

 

[Nick Shaw]

I think to improve the view of exposure it would be good to have:
1. A high / low key toggle. Where gamma is adjusted up/down, since sliding the iso is not the same since iso is a linear gain and not showing the DR in the same way as a gamma correction.

I don't deny that a hi/lo key toggle might be a useful function to be able to assign to a button.

However I am not sure that the Sony version is a gamma adjustment. The F65 manual is not very clear. It refers simply to "gain reduction" and "gain amplification", and if this is applied to the linear sensor data before the LUT, then it is doing exactly the same as an ISO adjustment.