EPIC MX Highlight Rolloff thread....

[Brian Boyer]

I'm not an engineer, so bear with me. A pixel receives a charge and returns a value that designates a level of luminance and hue. That value is either absolute white (what some are calling "no data" or 0 hue, max luma) or any hue at a luma level of -1, -2 etc..

What benefit should we expect to reap by artificially limiting the system's response so that it is unable to return a value of maximum luma? That is essentially limiting the DR of the sensor, isn't it?

Why cripple the system when you can basically chose to do the same thing yourself each and every time you set exposure?

I suppose it would be a rather simple feature request to implement. Ask away... "Dear Red, there are one too many crayons in your box. Kindly remove the whitest one. Please castrate my camera. Thank You."

OK, that was cheeky humor, not meant to offend. I know very little about uploading looks to the camera so I will ask. Is it possible to do what Brian is asking via a look?

Is it crippling when the part you're throwing away isn't captured data or actual picture information? I guess it depends on if the sensor overloads (a traffic light comes on) at the maximum luma value or anything beyond it. If it's the latter, what would you be losing?

If, for example, there are 1024 available values (0-1023), would overload begin at 1023 or 1024?

 

[Dave Blackham]

Its funny because Red and Alexa share the same sensor (only use it in different ways, from what I read, Red takes much of the sensor's power to capture more than 2Ks, whilst Alexa gave that power to DR and color rendition). But I'm not entirely sure this is at all true.

This is not correct. Red Designed their own sensor, Arri bought theirs in. They are not the same.

http://investors.cypress.com/releasedetail.cfm?ReleaseID=537967

If Im wrong then some one please correct me.

Id also add there's a lot more to it than just the sensor.

 

[Gunleik Groven]

Hi again.

have just been skimming through, but I think it could be good to set up some definitions.
Here's a shot:

"Clip"
What happens when a digital recording device gets more signal than it can handle. Result is digital distortion.
What you can do to avoid clip, is to apply an anlogue limiter in the signal chain.
For digital imaging devices, that would be aperture, shutter, ND, diffusion or all the above.
You cannot avoid "clip" on any digital signalrecording device if the signal is hotter than what the device is capable of receiving.
Different devices filter the resulting noise differently.

When recording images, it is important to remember that you actually record three channels of signal, and depending on the light-conditions, one chanel can clip before the other two. You will then get distortion in your image, even though the average lumavalue is lower than what the sensor should clip at.
A common example is that the red channel can clip under tungsten. This is espcially common if you wb for tungsten and shoot without headroom for signalprocessing.
In some occations, the WB process itself can introduce clipping-like symptoms in the red channel. For those occations, it is better to do the whiteballance with general color tools (like resolve) instead of the WB slider.

In imaging devices, the manufacturer try to filter clipped signals to white, as that is what we would usually/mostly see them as.
You cannot resolve "real" data from the noise.

Now next:
White/grey

A signal with even intensity in all three channels.
It is no distortion introduced to create this white signal, and if there is image-info hidden in that white signal, it can be recovered


Now:
When would clip be good?

I suggest that clip is good if you really do NOT want to recover signal in that part of the image (as underexposure can be good for the same reason. It's BLACK dammit!

Another good reason to clip, is because what you really want to expose for gets underexposed, in case you expose for the extreme highlights.
Sun comming into an un-ND'ed window with a person standing in or besides that window, or sun through branches in a dark forest, are pretty common reasons for clipping.
AND
It wasn't bad on ilm, and fortunately now it isn't bad in digital either.

Now.
Next?

 

[David Battistella]

I just said that if he shot on a Red, he wouldn't be able to work the same way. He would either have to spend more of his attention to look after the camera's limits, or he would light differently.

Right, and this whole thread has been about how to work with those tools,how to work with it and light it differently.

I think you are moving it more toward sensor design and wondering what RED could do to redesign their sensor to perform according to what you percieve as a better design in the Alexa. You are seemingly triggered particularly by the fact that Mark Toias dragon test does not satisfy what you wanted or were expecting from a new RED sensor design and you have expressed your disappointment.

Maybe the wrong questions are being asked? Maybe more needs to be understood about the choices and decisions that go into how a sensor performs under very specific lighting conditions. It's quite a technical discussion, where this thread is endevouring to resolve and show specific examples of how to handle these types of lighting situations with this sensor.

Hopefully, it has helped with regard to best practices in these situations with the current RED EPIC MX sensor, the only one presently available for the public to make films with.

It's not that I am not following the discussion. It's sort of a question of what choices are made in sensor design and that is not my realm. I'm not saying that your points are not valid. But I can not answer questions about sensor design choices. It's not that I am not curious about that myself, but I have to work with what I have, I can can wish for a lot more (and I think Dragon is a lot more based on what's have seen) but i am trying to stay in the realm of the tactile in this discussion.

Maybe another thread could be started on the merits of sensor design and maybe some of the REd team will be more than willing to engage in that conversation. I also understand the interconnected nature of all of this thinking, but it can very easily lead to disengament because the subject matter can be dense and is not within the realm of the cinematographer but the scientists who are creating the imaging system.

the answers to your types of questions need to be answered by the designers of the technology, not so much the users of it.

Its late in my Part of the world. I'll look at this thread again tomorrow.

david

PS: when any sentence starts with with the words "you people", it implies something, that something may not always be received in the most positive way.

 

[Scott Crawley]

Is it crippling when the part you're throwing away isn't captured data or actual picture information? I guess it depends on if the sensor overloads (a traffic light comes on) at the maximum luma value or anything beyond it. If it's the latter, what would you be losing?

It is data. No color, max luma is data, no? It just so happens that when you give that data to a grading system and ask for luma -1 it can give you grey, but chroma -1 is essentially meaningless. That seems to be why some think of it as no data... Because it contains no chroma information that can be accurately manipulated. Is that about right? (Sorry, not a Colorist either.) Who knows color sampling theory and can translate what I am trying to say into real terms, if possible?

What is beyond maximum luma? Maximum is the end of the line by definition.

 

[Björn Benckert]

yes, DPs want to overexpose for artistic reasons and they have to do it in a controlled way then it has to be because it needs to be overexposed in a way that relates to the rest of the shot or scene, not because your camera can't handle it.

i read somewhere that no other than Roger Deakins when overexposing the sky or other such glaring highlight he doesn't even read the highlights value, he doesn't care where those highlights values are, he just wants them overexposed in relation to the rest of the scene. This is how many great DPs work, they don't want to work for the camera they want the camera to work for them. If Roger choose to shoot on the Red, he wouldn't be able to work the same way, because Red will not control data overload for him. And if you are famiar with his work, you'll know that he would never allow for a clipped highlight in his film. There is not one bad technical or artistic shot in that man's career!

Same with Jeff Cronenworth, his Red shot movies are perfection. But his lighting style is perfect for Red. He's a master of soft light. He even worked as an assistant for Sven Nykvist, the father if soft light.

But hard lighting is a different beast altogether... And I believe Red is not the best option for a hard light film. At least not until they choose to address that problem.

And yes Luis I understand that the clipping gets worse when cir space is transferred to sEGB, but Red is not the only camera that has to go through that conversion, every camera has, and this problem is much more evident on the Red footage.

That brings me to a whole other issue, I can't wait for the day when Red decides to make their color space ProPhoto instead of RGB. Only then will things really get interesting!

I think you are way off,
As the red rolloff is beautiful. And I think Deakins would agree.

If you got uggly clipping highlights you over exposed your shots and / or do not know how to grade.

When I write overexposed your shot I do not mean that you just clipped the image. With overexposure I mean that the cliping point was far bellow were you wanted it to be in your finak grade. Basically if you shoot so that you can bring up the near clipping values without getting brighter than you intended. You will see that the clipping was not an issue. However if you shoot and expose in a way so that you in post want to bring down your near clipping values by a lot then you will fur sure have ugliness around the clipping point.

So again it's all about how you shoot, and how you expose. If exposed correctly it will look beautiful, if not it will look like shit, that pretty much goes for all cameras as no camera is fail safe in that regard, not the alexa not the dragon and not film. They all have a DR you need to keep what you want out of the picture within that DR or it will look ugly it's that simple.

Nice rolloffs can be made in camera but it's actually a grade thing, and it's quite complex. 3Dluts manage it quite well, a normal grading tool where you pull curves around does not.

 

[António Fagundes]

Now:
When would clip be good?

I suggest that clip is good if you really do NOT want to recover signal in that part of the image (as underexposure can be good for the same reason. It's BLACK dammit!

Another good reason to clip, is because what you really want to expose for gets underexposed, in case you expose for the extreme highlights.
Sun comming into an un-ND'ed window with a person standing in or besides that window, or sun through branches in a dark forest, are pretty common reasons for clipping.
AND
It wasn't bad on ilm, and fortunately now it isn't bad in digital either.

Now.
Next?

I really don't want to be indelicate but this doesn't make any sense.

 

[António Fagundes]

I think you are way off,
As the red rolloff is beautiful. And I think Deakins would agree.

If you got uggly clipping highlights you over exposed your shots and / or do not know how to grade.

When I write overexposed your shot I do not mean that you just clipped the image. With overexposure I mean that the cliping point was far bellow were you wanted it to be in your finak grade. Basically if you shoot so that you can bring up the near clipping values without getting brighter than you intended. You will see that the clipping was not an issue. However if you shoot and expose in a way so that you in post want to bring down your near clipping values by a lot then you will fur sure have ugliness around the clipping point.

So again it's all about how you shoot, and how you expose. If exposed correctly it will look beautiful, if not it will look like shit, that pretty much goes for all cameras as no camera is fail safe in that regard, not the alexa not the dragon and not film. They all have a DR you need to keep what you want out of the picture within that DR or it will look ugly it's that simple.

Nice rolloffs can be made in camera but it's actually a grade thing, and it's quite complex. 3Dluts manage it quite well, a normal grading tool where you pull curves around does not.

Bjorn, sorry but you're the one who's way off. Make the effort to read past posts, we're way past that argument. It's not about technical ability, it's about sensor design.

 

[António Fagundes]

Right, and this whole thread has been about how to work with those tools,how to work with it and light it differently.

I think you are moving it more toward sensor design and wondering what RED could do to redesign their sensor to perform according to what you percieve as a better design in the Alexa. You are seemingly triggered particularly by the fact that Mark Toias dragon test does not satisfy what you wanted or were expecting from a new RED sensor design and you have expressed your disappointment.

Maybe the wrong questions are being asked? Maybe more needs to be understood about the choices and decisions that go into how a sensor performs under very specific lighting conditions. It's quite a technical discussion, where this thread is endevouring to resolve and show specific examples of how to handle these types of lighting situations with this sensor.

Hopefully, it has helped with regard to best practices in these situations with the current RED EPIC MX sensor, the only one presently available for the public to make films with.

It's not that I am not following the discussion. It's sort of a question of what choices are made in sensor design and that is not my realm. I'm not saying that your points are not valid. But I can not answer questions about sensor design choices. It's not that I am not curious about that myself, but I have to work with what I have, I can can wish for a lot more (and I think Dragon is a lot more based on what's have seen) but i am trying to stay in the realm of the tactile in this discussion.

Maybe another thread could be started on the merits of sensor design and maybe some of the REd team will be more than willing to engage in that conversation. I also understand the interconnected nature of all of this thinking, but it can very easily lead to disengament because the subject matter can be dense and is not within the real of them cinematographer but the scientists who are creating the imaging system.

the answers to your types of questions need to be answered by the designers of the technology, not so much the users of it.

Its late in my Part of the world. I'll look at this thread again tomorrow.

david

PS: when any sentence starts with with the words "you people", it implies something, that something may not always be received in the most positive way.

David you are right and that's why I don't think I have anymore to add to this thread, as will obviously turn out to be about the merits of the MX and not about what could be improved.

And I sincerely apologize if I came out as rude, it wasn't my intention at all. But your friend was in fact extremely rude, and it gets exhausting to try and maintain these threads relevant to information when there are people like your friend that don't have manners and will try to bash anyone in a blink of an eye. And I just can't put up with that.

 

[Brian Boyer]

It is data. No color, max luma is data, no? It just so happens that when you give that data to a grading system and ask for luma -1 it can give you grey, but chroma -1 is essentially meaningless. That seems to be why some think of it as no data... Because it contains no chroma information that can be accurately manipulated. Is that about right?

What is beyond maximum luma? Maximum is the end of the line by definition.

I never said it wasn't data. I said it wasn't captured data from the scene in question - that's what I meant by picture information. Does that change your answer any?

And, does anyone know if the overloaded value is inclusive of the last value in the range or does it lay outside?

It may not be technically possible, practical or feasible to do what I'm inquiring about but there would seem to be little lost by not having to manage data that's not actually from the scene.

And to be clear, I'm not advocating camera and/or sensor manufacturers do this, I was just curious as to why they might not. I thought it was a good question and it was a shame that people seemingly weren't recognizing it among the other focus of this thread.

EDIT: Scott, I appreciate you at least trying to tackle it. Maybe no one else did because they thought it was off topic but I thought it carried over from what Antonio was trying to get at in the other thread as well, but got buried in the mix, so to speak.

 

[Scott Crawley]

I never said it wasn't data. I said it wasn't captured data from the scene in question - that's what I meant by picture information. Does that change your answer any?

Not really. It doesn't have chroma information, but the luma is useful to me if for nothing other than a reference point equal to "really fucking bright" Ya know? Much beyond this is way out of my depth. I'm reaching for straws already. ;-)

 

[Luis Otero]

...His point is, if he WANTS the appearance of blown out portions of the frame where everything is white (like in Gunleik's example), why do cameras have to overload the sensor in order to do it? Why can't they stop just short of the overload value and just give an actual white pixel if everything beyond it is supposed to be white anyway? ...

Brian,

I think that the answered to that specific question has already been provided and discussed here. The camera does not have to stop short of the overload value. This is why: if after you acquire the footage using Red (or even during shooting you set your monitor to view it as how the settings calls it RAW) if you set to view it and grade from RedLog Film/RedColor3 setting, you will see that the overexposed and/or clipping spots will appear just as it will be displayed by Cineon: a Log white. In other words, if you shoot the same scene with an equivalent film stock and scan it for DI grading, both will look the same.

Why? Again, because that combination of setting will present your RAW image from the R3D as a 100% equivalent to Cineon. It does not have any color artifact burned-in, as the magenta situation discussed before on this thread, and the roll-off will look as Cineon will look; both contain images in a Log fashion. Any other color/gamma selection will provide colors interpreted by an algorithm. Sometimes a well shot scene looks on RC3/RG3 exactly how you wanted it. Even Mark said that he used some of the footage on his test as they came out of camera, but that the majority of the footage came from RLF/RC3 with a simple curve he applied on RCX.

As I discussed above, I think that the artifacts are just a by product due to Color Space conversion, which was needed since it will be posted on Vimeo. BTW, that is what Vimeo is doing after the uploading is completed and making it ready for viewing (or as they call it, optimizing it). However, the RAW will not have any of those artifact.

I hope this helps,

 

[Brian Boyer]

"Clip"
What happens when a digital recording device gets more signal than it can handle. Result is digital distortion.

SNIP

You cannot avoid "clip" on any digital signalrecording device if the signal is hotter than what the device is capable of receiving.

But you could "clip" the sensor and just lop off the values that aren't adding to the image you're capturing, couldn't you? Again, it seems like it depends on whether or not the sensor "clips" at the max luma level or beyond it. Maybe overfill the bucket and just use (record) what's left inside, disregarding the overflow?

Is it more trouble than it's worth to have an option where everything that's compressed and recorded falls within the sensor's capabilities and everything that doesn't gets thrown out before it gets to REDCODE?

 

[Brian Boyer]

Not really. It doesn't have chroma information, but the luma is useful to me if for nothing other than a reference point equal to "really fucking bright" Ya know?

This makes me wonder if "really fucking bright" changes much, if at all, once you've overloaded the sensor. Can it go from "really fucking bright" to "really fucking brighter"?

In other words, assuming the brightest luma value the sensor is capable of producing is captured before the overload begins, are you really getting any more luma info at that point or is it fixed? I guess the same question applies even if the brightest value the camera records is the already overloaded (clipped) value.

I'm reaching for straws already

Word. Me, too...obviously.

 

[António Fagundes]

Someone was already on this... back in 2000!



From, SARITH: Safe ARITHmetic - A Progress Report


'Saturation ARITHmetic is applicable in any area of the problem domain where the size of numerical representation is critical, the limit of numerical representation is conceivably within reach, and where a closest representable value in the result is preferable to a "wrap-around" result. It is certainly not new: Ada is an example of a high-level language that implements saturation; and a good example of saturation support in hardware comes with MMX...

If, say, for an 8-bit example, p is signed 0x99 (a negative hex value) and q is unsigned 0x99 (a larger, positive hex value), p = q returns False, and p < q returns True.

Safe ARITHmetic is the natural complement to saturation arithmetic. It is applicable most typically (though not always) to situations of the problem domain where "an overflow IS a bug"; --i.e.: where we want to *detect* rather than "adjust for" overflows...

A future goal of SARITH is to bring MMX support to the Eiffel language. Unlike previous (inelegant) attempts to bring SIMD support to high level languages, SARITH aims to first support saturation and other behaviours with non-SIMD types, so as to "pave a road" to MMX in particular, and to SIMD more generally.'


So, 12 years later there must be something out there that can manage computing data overload?


From Wikipedia:


'In digital signal processing, clipping occurs when the signal is restricted by the range of a chosen representation. For example in a system using 16-bit signedintegers, 32767 is the largest positive value that can be represented, and if during processing the amplitude of the signal is doubled, sample values of 32000 should become 64000, but instead they are truncated to the maximum, 32767. Clipping is preferable to the alternative in digital systems — wrapping — which occurs if the digital hardware is allowed to "overflow", ignoring the most significant bits of the magnitude, and sometimes even the sign of the sample value, resulting in gross distortion of the signal.

The incidence of clipping may be greatly reduced by using floating point numbers instead of integers. However, floating point numbers are usually less efficient to use, sometimes result in a loss of precision, and they can still clip if a number is extremely large or small.'

 

[Scott Crawley]

This makes me wonder if "really fucking bright" changes much, if at all, once you've overloaded the sensor. Can it go from "really fucking bright" to "really fucking brighter"?

Dragon does. =-) relative to the MX, I mean

 

[Brian Boyer]

Brian,

I think that the answered to that specific question has already been provided and discussed here. The camera does not have to stop short of the overload value. This is why: if after you acquire the footage using Red (or even during shooting you set your monitor to view it as how the settings calls it RAW) if you set to view it and grade from RedLog Film/RedColor3 setting, you will see that the overexposed and/or clipping spots will appear just as it will be displayed by Cineon: a Log white. In other words, if you shoot the same scene with an equivalent film stock and scan it for DI grading, both will look the same.

Why? Again, because that combination of setting will present your RAW image from the R3D as a 100% equivalent to Cineon. It does not have any color artifact burned-in, as the magenta situation discussed before on this thread, and the roll-off will look as Cineon will look; both contain images in a Log fashion. Any other color/gamma selection will provide colors interpreted by an algorithm. Sometimes a well shot scene looks on RC3/RG3 exactly how you wanted it. Even Mark said that he used some of the footage on his test as they came out of camera, but that the majority of the footage came from RLF/RC3 with a simple curve he applied on RCX.

As I discussed above, I think that the artifacts are just a by product due to Color Space conversion, which was needed since it will be posted on Vimeo. BTW, that is what Vimeo is doing after the uploading is completed and making it ready for viewing (or as they call it, optimizing it). However, the RAW will not have any of those artifact.

I hope this helps,

I didn't comment on magenta color casts in RED images so I kind of glanced over comments others may have made about them, as well as the direct responses to them, so I admit I probably missed their relevance to this discussion.

Are you saying that it's really as simple as any color artifacts seen in footage are not actually there but are really only the result of the various color space algorithms? Starting from REDlogfilm will give a truer representation of the actual scene without introducing unwanted color shifts?

I always start with REDlogfilm so, if that's true, I've been dodging a potential bullet this whole time without knowing it.

It's probably helpful to mention I haven't really had any problems with artifacting so I'm not coming at this topic from that angle. Like I said elsewhere, I just wanted to help Antonio get an answer to what I thought was a good question.

I'll chew on everything everyone wrote and give this topic a rest.

 

[Luis Otero]

The most important fact to remember, is that there is no final Color Science for the Dragon sensor, yet. This is just a test drive of an unfinished endeavor in which Red has continued to show its eagerness of keep transparency.

Even with those limitations, we got Mark to deliver a superb mix of footage that exceeds any one in the market, and by his own words, even 65mm film. With an unfinished product...

Knowing what Red's Color Science chef has been capable to do in the past (at this point, he does not have to prove himself to any one), what we are witnessing is something that never has been able to achieve in such a small amount of time.

Looking forward to see more blockbuster movies shot with a camera half the size of a tin lunchbox...

 

[Luis Otero]

Dragon does. =-) relative to the MX, I mean

Love it, man...!

 

[Bill Anderson]

If Mark's image had been accompanied by a sample from another camera, and that sample had shown the same low and mid values while it held good separation in the highlights, that would have been something worth pointing out as far as the Dragon image is concerned. But that is not the case, and given the specs of Dragon, such a camera might not exist! There is absolutely no telling how far Mark pushed the highlight values up the scale and into clipping in order to place his lower and mid values, and there is no way to know just what highlight separation there was in the sky! So why point your finger at a place with so few known coordinates and say this is where it's at.

Discussing the nature of clipping is all good though. No doubt about that.