fredmiranda.com Login
This topic is in the 2010 archive. You can read it and like its posts here. It is read only.

Archive 2010Sigma SD1

  
 
theSuede
Offline
Details
p.13 #1Sigma SD1
So.... Let's say that you've got a RED Ferrari, and the perfect RED-haired model to go with it. You get her a RED dress, and a RED set of shiny stiletto heels to match. You've spent two full days getting ALL of the reds in the scene to match PERFECTLY. They seem to be IDENTICAL to the human eye, at least in normal sunlight (which is what we are shooting in, sunlight and passive reflectors!). You take the shots, go home and download the pictures.

Now the reds don't match anymore (?!)
The car is slightly orange and too bright, the shoes are a lot darker than you remembered them, but the right hue - and the dress is a mess of too cold/warm hues of red depending on the light/shadow on the part of the fabric you're looking at. The red hair is also too dark.

HOW are you going to solve that with a LUT? Simple answer - you can't!

The reason for this is metamerism failures. Your eyes record colour in a very special way, and in THIS way all the reds actually seemed to be identical - even though their reflected light spectras are all different. It doesn't matter, your eyes registered all those mixes as "identical". Same hue, same saturation, same lightness. To get a camera to do the same the filters in the camera - the camera "fingerprint" of colours - must match human vision at least up to a certain point. If they don't, then you'll be able to correct ONE of the reds to match what you saw in reality. But since the other reds aren't made from the same material they will be rendered as if they were a totally different colour.

The principle behind the Foveon sensor makes metamerism conformance with human sight impossible. The best you can get is a minimum 6% volumetric error - which is quite a lot. Most Bayer-based cameras manage about 0.2-1%.

This is only a real deal-breaker in some certain cases like fashion (make-up and fabric colour is a b*tch with Foveons) and maybe most of all - product photography. Often products are made of several different materials that are visually matched to to give the "same colour" impression, something the Foveon doesn't do very well. And now we're only discussing the best kind of light there is - sunlight.
If you take a more difficult light like fluorescents or streetlights the Foveon cannot under ANY circumstance be made to mimic human vision. What you get is something entirely else. It might look good anyway, but it will NOT be what you saw.

In a street/landscape shot this might be less important. So it's not a total "fail", it just limits the camera somewhat. Unfortunately the amplified noise signature and weak overall colour response limits it even further. So, yes - you get a camera with a very high colour resolution (spatial resolution, not hue...). Does that really outweigh the downsides? In a consumer/capitalistic oriented society, people vote with their wallets. And the vote is still out. But the prognosis doesn't look too good.
Oct 05, 2010 at 08:04 PM
Tariq Gibran
Offline
Details
p.13 #2Sigma SD1
theSuede wrote:
So.... Let's say that you've got a RED Ferrari, and the perfect RED-haired model to go with it. You get her a RED dress, and a RED set of shiny stiletto heels to match. You've spent two full days getting ALL of the reds in the scene to match PERFECTLY. They seem to be IDENTICAL to the human eye, at least in normal sunlight (which is what we are shooting in, sunlight and passive reflectors!). You take the shots, go home and download the pictures.

Now the reds don't match anymore (?!)
The car is slightly orange and too bright, the shoes
...Show more →

From what I have seen recently, SPP 4 and the True II processor seem to have taken care of the red issue. I have seen a number of Sigma shots which contain different shades of red that reproduce accurately. Check out the shot of the model with red hair and red lipstick below.

http://vstudio-international.com/blog/vstudio/fluid-females/
Oct 05, 2010 at 08:48 PM
theSuede
Offline
Details
p.13 #3Sigma SD1
?
This is not something you can "take care of" with software, it's a pure physical fact. Colours that look the same in reality may look as two different colours in the files. Unless you're the person behind the linked vstudio shots, you actually have no idea what those reds looked like in reality. For all we know (and knowing it's shot with a Sigma), the reds could be inverted. The hair in the "red" shot is warmer in hue than the lipstick and the fluid. Chances are about 50/50 that the other way around was what was intended (and also what it looked like). This makes it pretty hard on both the make-up artists and the photographer. Spray & pray. It MIGHT look ok, it might not. Those two colours that you matched MAY look the same in the file, and they may not. You'll never know until you look at the files.

The problem here isn't "making a shot look nice", it's when you actually know what you want the shot to look like, or have to match colours.

Another more related situation would be shooting sports. Let's say you follow a team, and they do both indoor and outdoor games. Mid-day and evening games. With some cameras you can make the uniform colours AND the skintones AND the cheerleading team colourmatch perfectly from game to game in the year-book no matter where or when the shot was taken - and with some cameras that will be impossible (unless you do manual colour correction with masking).
Oct 05, 2010 at 09:37 PM
Tariq Gibran
Offline
Details
p.13 #4Sigma SD1
theSuede wrote:
?
This is not something you can "take care of" with software, it's a pure physical fact.


The True II is an image processing hardware chip within the camera. I do think it's a bit premature to be stating that the new Foveon implementation used in the SD1 is physically incapable of reproducing color accurately. That's something we will just have to see. Hopefully the ad for the SD1 will feature a red head eating strawberries on the hood of a red Ferrari.
Oct 05, 2010 at 10:30 PM
Yakim Peled
Offline
Details
p.13 #5Sigma SD1
Tariq Gibran wrote:
theSuede wroteSo.... Let's say that you've got a RED Ferrari, and the perfect RED-haired model to go with it. You get her a RED dress, and a RED set of shiny stiletto heels to match. You've spent two full days getting ALL of the reds in the scene to match PERFECTLY. They seem to be IDENTICAL to the human eye, at least in normal sunlight (which is what we are shooting in, sunlight and passive reflectors!). You take the shots, go home and download the pictures.

Now the reds don't match anymore (?!)
The car is slightly orange and too bright, the shoes are a lot darker than you remembered them, but the right hue - and the dress is a mess of too cold/warm hues of red depending on the light/shadow on the part of the fabric you're looking at. The red hair is also too dark.

HOW are you going to solve that with a LUT? Simple answer - you can't!

The reason for this is metamerism failures. Your eyes record colour in a very special way, and in THIS way all the reds actually seemed to be identical - even though their reflected light spectras are all different. It doesn't matter, your eyes registered all those mixes as "identical". Same hue, same saturation, same lightness. To get a camera to do the same the filters in the camera - the camera "fingerprint" of colours - must match human vision at least up to a certain point. If they don't, then you'll be able to correct ONE of the reds to match what you saw in reality. But since the other reds aren't made from the same material they will be rendered as if they were a totally different colour.

The principle behind the Foveon sensor makes metamerism conformance with human sight impossible. The best you can get is a minimum 6% volumetric error - which is quite a lot. Most Bayer-based cameras manage about 0.2-1%.

This is only a real deal-breaker in some certain cases like fashion (make-up and fabric colour is a b*tch with Foveons) and maybe most of all - product photography. Often products are made of several different materials that are visually matched to to give the "same colour" impression, something the Foveon doesn't do very well. And now we're only discussing the best kind of light there is - sunlight.
If you take a more difficult light like fluorescents or streetlights the Foveon cannot under ANY circumstance be made to mimic human vision. What you get is something entirely else. It might look good anyway, but it will NOT be what you saw.

In a street/landscape shot this might be less important. So it's not a total "fail", it just limits the camera somewhat. Unfortunately the amplified noise signature and weak overall colour response limits it even further. So, yes - you get a camera with a very high colour resolution (spatial resolution, not hue...). Does that really outweigh the downsides? In a consumer/capitalistic oriented society, people vote with their wallets. And the vote is still out. But the prognosis doesn't look too good.
From what I have seen recently, SPP 4 and the True II processor seem to have taken care of the red issue. I have seen a number of Sigma shots which contain different shades of red that reproduce accurately. Check out the shot of the model with red hair and red lipstick below.

http://vstudio-international.com/blog/vstudio/fluid-females/


We had a bold idea for this series with clothing made out of liquids; water, milk, coffee and wine.

I'd dress that model above with red wine. I'm not sure the above description is complicated enough. But seriously, I'd really like to see a direct comparison under controlled environment.

Happy shooting,
Yakim.
Oct 06, 2010 at 01:19 AM
Lotusm50
Offline
Details
p.13 #6Sigma SD1
Tariq Gibran wrote:
The True II is an image processing hardware chip within the camera. I do think it's a bit premature to be stating that the new Foveon implementation used in the SD1 is physically incapable of reproducing color accurately. That's something we will just have to see. Hopefully the ad for the SD1 will feature a red head eating strawberries on the hood of a red Ferrari.



Of course if you read the Sigma/Foveon promotional materials, they are the ONLY one hoe can reproduce color accurately! This, of course goes to what I was saying before. You won't hear everything in the selected bits of oversimplified (so the punters can understand) science of the company materials.

Yes, I agree it is premature to judge the SD1. As described above, the camera has serious challenges due to the physical limitations of the design. It is probably true that the camera is indeed incapable of reproducing color accurately -- I don't doubt it. But how accurate is accurate, and how accurate does something have to be before it causes real problems in practice? Bayer sensors aren't perfect either (and neither is film), but both produce quite acceptable output. An important factor is how well the camera/software software deals with these challenges. When these challenges bind, how well or how effectively does the software fudge, interpolate, approximate, smooth, calculate, guess, etc. to produce acceptable output? It may turn out that the output is off, but it really isn't too bad. So, we do have to wait and see the overall implementation before we can judge. So which are more important -- the advantages Sigma and crew persistently trumpet, or the limitations they are silent about? What is clear is that the Foveon sensor isn't magic bullet, perfect sensor that Sigma would like you to believe, but it could still be quite usable. We'll see.

Oct 06, 2010 at 08:01 AM
Tariq Gibran
Offline
Details
p.13 #7Sigma SD1
The other issue for Sigma is that they have historically played catch up to everyone else with regard to resolution. Let's say for arguments sake that the SD1 is equivalent to a 30MP Sony, Nikon or Canon bayer sensored camera. Does anyone really believe that we will not see 30+ MP FF cameras around the 2-2.5K price within the next year? My bet is that we will.
Oct 06, 2010 at 08:13 AM
theSuede
Offline
Details
p.13 #8Sigma SD1
Tariq - implementation in "hardware" via firmware programming or doing it in post is the same thing. It's just two different ways of programming. One is a program you run in a normal computer, one is a program you run in a specialized processor in the camera. Both are just - programs.

And this is a purely physical problem, not solvable by mathematical trickery or correction algorithms. It's a math problem with six unknowns and a skewed measurement and an unknown result target (phew!) - and even IF you solve all the parts of the problem correctly you will STILL be physically unable to recreate the colours you, as a human, see in the scene...

The difference is in the way the camera measures colour. And it's not like it's a problem isolated to red hues, it's there in all hues.
http://ezpicshare.com/images/foveoncomp.gif
The response from the Foveon sensor differs on so many points (local maxima centers, slopes, cross-over-points and so on) that no matter what you do, colours in the file will NOT look like you see them in reality.

You can always use software in- or out of camera to make A FEW spot colours match reality. But hues that look very similar - or even identical - to them might be rendered as something totally different. This is not correctable. You just have to live with it.

For some people this matters, for some it doesn't. But from a purely objective point of view, why would you want a camera that is guaranteed to never see reality as you do? Aren't cameras meant to mirror reality, giving us a picture that we later can process to look like whatever we want them to?

Me, I like the base material (my raw files) to reflect reality as well as it is possible. Even in the cases when I process the crap out of them. Having a good similarity to reality gives me the best possible starting point when going to PP.
Oct 06, 2010 at 05:29 PM
U.C.
Offline
Details
p.13 #9Sigma SD1
Well, I think you're right about colour-reproduction. It's very preferable that a camera reproduces the same image as seen by the human eye. But on the other side, some of us just like the things the human eye can't see. I'm thinking about (very) shallow depth of field with f/1.2 lenses (which is almost "common" at the alt-board), or the ability to "gain" light in a scene (iso6400 + f/1.2 delivers more light than we can see). Camera's are also capable to capture (much) more detail in a single shot than we can see.
But, I know, the camera should at least be capable to produce comparable levels of colour-reproduction, DoF, light, sharpness, contrast, dynamic range as the human eye. I think we have to wait some time for camera's capable of having the DR of the human eye.
Oct 06, 2010 at 06:04 PM
Tariq Gibran
Offline
Details
p.13 #10Sigma SD1
It's very difficult for me to wrap my brain around the idea that a black and white sensor with color filters is capable of accurate color while a true color sensor is not. It may be the case, I don't know but it just seems logical that if accurate color is possible form a black and white sensor utilizing whatever tricks then the same must be possible with the Foveon sensor.

One of the main jobs I do with my DSLR is to make highly color critical reproductions of original art work. If I do go down the SD1 road, you can bet I will buy from Amazon or a similar place that allows returns should the camera be incapable of accurate color reproduction as tested under 5500K UV corrected strobes in the studio.
Oct 06, 2010 at 06:15 PM
Daniel Heineck
Offline
Details
p.13 #11Sigma SD1
It's a different method of color measurement, that's why. With the Bayer CFA, the color response can be tuned very carefully because its response is not completely tied to the absorption depth characteristics of silicon. In other words, if we made a foveon-based sensor out of GaAs/AlGaAs or Ge or InP or whatever semiconductor material you like that day, it would have drastically different spectral response. With an externally applied CFA, there is another couple knobs to tune.

That's why. Hope that helps.
Oct 06, 2010 at 06:51 PM
Tariq Gibran
Offline
Details
p.13 #12Sigma SD1
Daniel Heineck wrote:
It's a different method of color measurement, that's why. With the Bayer CFA, the color response can be tuned very carefully because its response is not completely tied to the absorption depth characteristics of silicon. In other words, if we made a foveon-based sensor out of GaAs/AlGaAs or Ge or InP or whatever semiconductor material you like that day, it would have drastically different spectral response. With an externally applied CFA, there is another couple knobs to tune.

That's why. Hope that helps.


So are you saying that Sigma would not specify the exact same material with a known spectral response to ensure consistency? If the material has a known response that Sigma has engineered for, then they could also theoretically "tune" said response however they so need to. Since there have not been that many generations of Foveon sensors, it seems like with this new sensor, they have another opportunity to improve upon whatever color issues have existed in the past. I guess we will just have to see.
Oct 06, 2010 at 08:09 PM
kwalsh
Offline
Details
p.13 #13Sigma SD1
Tariq Gibran wrote:
If the material has a known response that Sigma has engineered for, then they could also theoretically "tune" said response however they so need to.


No, they can't, and that is the point. Foveon uses photon diffusion in the semiconductor itself to separate colors between the three photodiodes. While you can control where the colors cross-over you can not, in anyway, change the extremely broad spectral response of each channel. This is the very nature of diffusion, the physics always works out this way and no material "tuning" will narrow the response. Thus they are "stuck" with the more broad than desired responses of the three layers and this is where some of the color conversion issues can appear (again, emphasis on can, the process clearly works quite well but it has different benefits and issues than the CFA approach).

For a CFA the filtering process is not diffusion, but absorption from chromogenic dyes. In these cases changing the chemistry of the dyes allows one to tune the spectral response and narrow it as much as desired (specifically to better match human spectral responses to prevent metamerism). By choosing different dye chemistries one can on a very sharp spectral scale select which energy (color) photons are absorbed and which pass on to the photodiode for detection. Since this process is absorption, not diffusion, there are many, many, many degrees of freedom in choosing each channels spectral response. Not so with diffusion, there is really nothing to adjust that has any impact on the narrowness or broadness of the spectral response.

And please keep in mind, all photodiodes are "black and white". Nothing about the Foveon sensor makes it "color" any more than a Bayer sensor. Foveon has three monochrome sensors stacked on top of each other, the stacking performing the color filtering via diffusion. The Bayer has four monochrome sensors sitting next to each other with chromogenic filters on top of them. Just different physical arrangements of the same monochromatic elements with color filters in front of them. In the Foveon case the upper sensors actually act as the filters for the lower sensors, but they are still just monochromatic sensors with filters in front of them.

So Foveon gets to measure three colors at each location (good) but they are restricted in the choice of the color channel response (not so good).

How "bad" are the broad Foveon responses? That is really hard to say. Like everything else about Sigma/Foveon we see relatively slow development. Is a particular case of weird color a limitation of the broad spectral response or is just because they "aren't doing it right" just yet? Can't really say, but by many accounts color has been getting better over time with the same sensors and just newer RAW converters - so it would seem they have still been learning and improving how they work around the broad spectral responses. We can speculate all we want about the impacts. The reality though is Foveon has very little choice over the spectral response of their channels, they just have to deal with what they get to a large degree and handle any challenges in the RAW converter. CFA can't benefit from "full color" pixels and must demosaic but it can optimize the spectral responses to make the color job easier in the RAW converter. Which is better? I don't know, but they are definitely different.
Oct 06, 2010 at 09:59 PM
nugeny
Offline
Details
p.13 #14Sigma SD1
There has never been any device--- films, sensors,--- that can see things like human eyes see. Heck, we don't see things the same ways either. That is why we have arts to please every taste.
So if Foveon doesn't see things the way some ones see, it wouldn't be an disadvantage. For some it may be a source of creativity in art.
Oct 07, 2010 at 11:46 AM
Daniel Heineck
Offline
Details
p.13 #15Sigma SD1
nugeny wrote:
There has never been any device--- films, sensors,--- that can see things like human eyes see. Heck, we don't see things the same ways either. That is why we have arts to please every taste.
So if Foveon doesn't see things the way some ones see, it wouldn't be an disadvantage. For some it may be a source of creativity in art.


Sure there's a ton of eye-brain interaction "stuff" going on that will change how we perceive colors, but deep down inside your eye and mine work the same. This is what Joakim is talking about.

The photosensitive chemicals in your eyes have a very specific spectral responses and exist in tightly regulated quantities. Therefore, it's fairly well known what the eye "sees" color wise. That is the baseline information that the brain works with. I'm quite confident, regardless of the difference in artistic vision between the two of us, that you and I are using the same photosensitive chemistry in our respective eyes.

To what effect the difference in spectral response has on perceived metamerism is outside my knowledge, but this spectral difference is a physical/chemical phenomenon, not a perceptual one.

Ken--great writeup on the fundamental color difference between foveon and cfa-based systems. One super duper tiny nit is that photons don't really diffuse, but have statistical penetration depths based on energy (or wavelength, depending on how you wish to describe them) which serve to separate the colors. As I said, painfully small nit in an otherwise well written piece.
Oct 07, 2010 at 12:18 PM
nugeny
Offline
Details
p.13 #16Sigma SD1
Daniel Heineck wrote:
nugeny wroteThere has never been any device--- films, sensors,--- that can see things like human eyes see. Heck, we don't see things the same ways either. That is why we have arts to please every taste.
So if Foveon doesn't see things the way some ones see, it wouldn't be an disadvantage. For some it may be a source of creativity in art.
Sure there's a ton of eye-brain interaction "stuff" going on that will change how we perceive colors, but deep down inside your eye and mine work the same. This is what Joakim is talking about.

The photosensitive chemicals in your eyes have a very specific spectral responses and exist in tightly regulated quantities. Therefore, it's fairly well known what the eye "sees" color wise. That is the baseline information that the brain works with. I'm quite confident, regardless of the difference in artistic vision between the two of us, that you and I are using the same photosensitive chemistry in
...Show more →

"deep down inside your eye and mine work the same."

Supposed this statement is 100% true---It is debatable at best in reality--But for the simplicity of the argument, let's accept it as 100% accurate, it is just 50% of a equation of what and how we see. The other half, the brain is more important. We do see more with our brain than eyes.

Oct 07, 2010 at 01:43 PM
Daniel Heineck
Offline
Details
p.13 #17Sigma SD1
nugeny wrote:
Supposed this statement is 100% true---It is debatable at best in reality--But for the simplicity of the argument, let's accept it as 100% accurate, it is just 50% of a equation of what and how we see. The other half, the brain is more important. We do see more with our brain than eyes.



What is debatable about the physiochemical pathways that gives us our specific color sensitivity? I'd love to know--it's not color-blindness, that from what I understand is the quantity of receptors, not the response of each individual receptor (much the same as going to a rrgb instead of the usual rggb bayer, which doesn't change the spectral response, just color sensitivity). It's not the color receptor molecules themselves that change, unless there's some kind of major genetic defect. Those dye molecules are exquisitely specific, for sure. Correct me if I'm wrong.

The brain interprets said data, yes. But it has to do image reconstruction with what it gets from the eye much the same way that we post process what comes off a chip. And what we see may be 75% from the brain, I don't know. But I do know that it's based on the physical/chemical constraints of the eye.
Oct 07, 2010 at 02:06 PM
kwalsh
Offline
Details
p.13 #18Sigma SD1
Daniel Heineck wrote:
One super duper tiny nit is that photons don't really diffuse, but have statistical penetration depths based on energy (or wavelength, depending on how you wish to describe them) which serve to separate the colors. As I said, painfully small nit in an otherwise well written piece.


Exactly correct, it isn't actually diffusion, but the distribution shapes are similar and the term is sometimes (inappropriately) used in this context. I couldn't think of a more appropriate term to use at the time that wasn't more confusing. Diffusion is definitely the wrong word though, because there is a concept of "photon diffusion" for optically thick media which is entirely different.

Thanks for the clarification!

Ken
Oct 07, 2010 at 03:17 PM
nugeny
Offline
Details
p.13 #19Sigma SD1
Daniel Heineck wrote:
nugeny wroteSupposed this statement is 100% true---It is debatable at best in reality--But for the simplicity of the argument, let's accept it as 100% accurate, it is just 50% of a equation of what and how we see. The other half, the brain is more important. We do see more with our brain than eyes.
What is debatable about the physiochemical pathways that gives us our specific color sensitivity? I'd love to know--it's not color-blindness, that from what I understand is the quantity of receptors, not the response of each individual receptor (much the same as going to a rrgb instead of the usual rggb bayer, which doesn't change the spectral response, just color sensitivity). It's not the color receptor molecules themselves that change, unless there's some kind of major genetic defect. Those dye molecules are exquisitely specific, for sure. Correct me if I'm wrong.

The brain interprets said data, yes. But it has to
...Show more →

We are not clone of one another. Our eyes may have 99% in common, but this 1% make us see things differently from one another, there for we have so many different tastes. To make things more complicate, our eyes are under direct command of our brain that picks a chooses what and how to see and interprete. And just one more factor; We perceive same things differently at different times. NOt so rare that we change our opinions about the same thing, same situation.
Oct 07, 2010 at 04:20 PM
theSuede
Offline
Details
p.13 #20Sigma SD1
The EXACT hue that we see on a screen print will never match reality, because of white-point corrections, S-curve DR corrections and so on. This is not important - just as some have already pointed out.

But the other part of colour accuracy IS important. Our psychovisual impression of colour is based on RELATIVES, made up in the part of the cognitive cortex. Those RELATIVES describe how colours relate to eachother, if they are more/less saturated, lighter/darker, warmer/colder than some other absolute colour. They are based on the ABSOLUTES that the receptor pigments in our eyes measure. Our brain is the mother of all whitebalance algoritms, the ultimate reference for colour scientists.

So, that a "colour" on the screen isn't exactly matching the "absolute" you would see in reality is of less importance. This is secondary, a perception phenomena. Your brains' auto-WB will take care of it.

Now for the important part:

Even though we can "see" a certain absolute colour as different things depending on whitebalance and luminance and still perceive it as "the same colour", there's a very certain relation between different absolute colours. They keep their internal distances very constant, even though lighting and light temperature may change. THIS yellow is slightly more green than THAT yellow, and it's slightly more saturated. THIS red is slightly more saturated than THAT red, and slightly darker.

Those inter-relations between different colours are a lot more important than absolute colour. Global absolute colour shift can easily be corrected. Random inter-relation shifts cannot.

The Foveon principle makes those inter-relations impossible to track. The measurement that the sensor takes does not "translate" like the stimuli from your eyes, or a Bayer with appropriate filters. When you do the translations based on the Foveon response, you move those inter-relations around in a totally random way - depending on the spectral response of the original colour. And we don't know those.

This is what metamerism is all about. Two colours can look absolutely identical to your eyes, even if their spectras are totally different. A good camera also measures those two as the same colour. The Foveon does not.
Oct 07, 2010 at 04:30 PM
FM ForumsLeica & Alternative GearJoin Upload & Sell
  
 

You are not logged inLog inRegister