Canon 5D Mark IV brings dramatic dynamic range improvements to the 5D line

[KeithBreazeal]

Yeah, and the 5DIV is pretty close to the 1D X II (closer, I think, than the 5DIII vs. 1D X...but DPR never tested the 1D X).

Push the Nikon D5 images hard and they look like crap by comparison, but hey, DPR unbiasedly gave Nikon's flagship camera a pass on that one.

Word.
I spend a good amount of time comparing camera and ISO charts. I really don't care about the Nikon or Sony specs since I wouldn't buy them, but I did find it interesting how close the Mark IV was getting to the Sony compared to the Mark III. My interest is how much better the Mark IV is compared to the Mark III. The difference was huge. My decision to upgrade was made yesterday. My VISA card is still smoldering.
This forum is invaluable for research and comments. There are some really talented posters here and many with great in depth technical posts. For all those that posted all the great info, thank you.

 

[Mikehit]

but it seems to me that there should be a way to test where the sensor clips the highlights

The sensor clip the highlights because you overexpose as Aglet ably described. I'm not sure what your question really is.
Take the brightest part of the image, and set the camera so it does not clip. The question then comes how much of the image (ie shadow) is lost to unacceptable noise. In other words, what is the acceptable dynamic range.

What you seem to be talking about is when the manufacturer programs the metering system, what do they call 'normal exposure' compared to the highlights and that is a programming issue.

 

[rishi_sanyal]

Its interesting how the ability to recover highlights, or at least the useable highlight headroom is never tested or discussed.

Yes...that's because most discussions about DR are really about shadow pull using iso invariance. True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with and have a contrast scale beyond the legacy CCD capability of 8.4 stops. It's similar to slide film in it's DR, but way below the capabilities of negative print film's 14 stops of DR. The current fad with shadow recovery has little to do with DR. In the current Nikon and Sony sensors, yes you can pull shadows on a 100 iso image from apparent blackness to 5 stops of mid tones. But the equivalent iso noise is still 5 stops over the base iso 100. That's going to be iso 3200 noise in the shadows > new mid tones. That's really too high and it's not true HDR, it's just pulling noisy shadows using an iso trick.

"True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with" --> Same levels of ISO noise as what tone? Even for a camera with zero read noise, darker tones will have lower SNR, because of simple physics (shot noise, which you can read more about in our article here: http://bit.ly/shotnoise).

And remember: darker tones on a linear capture format (Raw) are only dark because your monitors are extremely dim. Take that same linear Raw file and display it on a bright HDR monitor, and you won't have to push any shadows to see them.

Does that mean those shadows on that HDR monitor will be clean, even if shot at ISO 100? If they're from the lower end of the Raw file, no, because those tones inherently have less SNR because of shot noise.

The only way you can make them cleaner is by capturing more total light, which means either (1) a larger sensor, or (2) higher full-well capacity. Medium format or larger might address 1, whereas technologies like ISO 64 on the Nikon D810, which extends full-well capacity, is another way to make those 'shadows' look cleaner.

Both approaches do so by raising SNR of all tones, but the dark tones will still have lower SNR than the brighter tones. But after a certain SNR threshold, you won't care.

Does that make sense? It's a bit of a complex topic so apologies if I haven't done a good job explaining here.

Rishi

 

[privatebydesign]

Its interesting how the ability to recover highlights, or at least the useable highlight headroom is never tested or discussed.

Yes...that's because most discussions about DR are really about shadow pull using iso invariance. True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with and have a contrast scale beyond the legacy CCD capability of 8.4 stops. It's similar to slide film in it's DR, but way below the capabilities of negative print film's 14 stops of DR. The current fad with shadow recovery has little to do with DR. In the current Nikon and Sony sensors, yes you can pull shadows on a 100 iso image from apparent blackness to 5 stops of mid tones. But the equivalent iso noise is still 5 stops over the base iso 100. That's going to be iso 3200 noise in the shadows > new mid tones. That's really too high and it's not true HDR, it's just pulling noisy shadows using an iso trick.

"True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with" --> Same levels of ISO noise as what tone? Even for a camera with zero read noise, darker tones will have lower SNR, because of simple physics (shot noise, which you can read more about in our article here: http://bit.ly/shotnoise).

And remember: darker tones on a linear capture format (Raw) are only dark because your monitors are extremely dim. Take that same linear Raw file and display it on a bright HDR monitor, and you won't have to push any shadows to see them.

Does that mean those shadows on that HDR monitor will be clean, even if shot at ISO 100? If they're from the lower end of the Raw file, no, because those tones inherently have less SNR because of shot noise.

The only way you can make them cleaner is by capturing more total light, which means either (1) a larger sensor, or (2) higher full-well capacity. Medium format or larger might address 1, whereas technologies like ISO 64 on the Nikon D810, which extends full-well capacity, is another way to make those 'shadows' look cleaner.

Both approaches do so by raising SNR of all tones, but the dark tones will still have lower SNR than the brighter tones. But after a certain SNR threshold, you won't care.

Does that make sense? It's a bit of a complex topic so apologies if I haven't done a good job explaining here.

Rishi

Rishi,

How is that 5DSR RAW file coming on?

 

[Lurker]

Premium camera in terms of MP, delivering less than premium IQ when compared to a "lesser" camera.

I think I see the flaw in the logic. You view the 5D IV as a lesser camera and 5D SR then as better camera. I don't think Canon views it that way. They intend to provide 2 equal cameras, thus both being 5D, with different strengths for different target groups. With some research you should be able to find the interview were Canon exec said they were going modular. They were not going to try and make 1 thing that met the needs of everyone. Too many people going in too many different directions for that.

Was it Henry Ford? You can have whatever color you want as long as it's black. Look were cars are today.

 

[RBC5]

Sometimes I wonder if Canon doesn't deliberately keep things subdued so that all the negative hype gets cleared out up front and then the reality of a good solid performance rises up out of that perspective and Canon gets the last laugh. I never though much about this until the D5 bragging about such high ISO settings.

When I stumbled upon this:

https://www.youtube.com/watch?v=IRqrvKdckys

I honestly couldn't believe my eyes. What morons would go to this guy for any advice on anything. ;D

Jack

My coworkers and I have been watching his videos for a month strictly for the entertainment value. The best ones are when he talks in a high voice and has imaginary conversations. Just hilarious. And then there is his reasoning for not buying a Sigma Art lens..."It's a turd." ;D

 

[Jack Douglas]

Sometimes I wonder if Canon doesn't deliberately keep things subdued so that all the negative hype gets cleared out up front and then the reality of a good solid performance rises up out of that perspective and Canon gets the last laugh. I never though much about this until the D5 bragging about such high ISO settings.

When I stumbled upon this:

https://www.youtube.com/watch?v=IRqrvKdckys

I honestly couldn't believe my eyes. What morons would go to this guy for any advice on anything. ;D

Jack

My coworkers and I have been watching his videos for a month strictly for the entertainment value. The best ones are when he talks in a high voice and has imaginary conversations. Just hilarious. And then there is his reasoning for not buying a Sigma Art lens..."It's a turd." ;D

Sorry for implying you guys are morons. Actually, I called my wife and we both watched it and had a good laugh so I guess that's what it's all about.

Jack

 

[dsnook]

but it seems to me that there should be a way to test where the sensor clips the highlights

The sensor clip the highlights because you overexpose as Aglet ably described. I'm not sure what your question really is.
Take the brightest part of the image, and set the camera so it does not clip. The question then comes how much of the image (ie shadow) is lost to unacceptable noise. In other words, what is the acceptable dynamic range.

What you seem to be talking about is when the manufacturer programs the metering system, what do they call 'normal exposure' compared to the highlights and that is a programming issue.

Thanks for the response. I think what you have described is exactly how DPR and others think about DR. My question is mostly why is this the way that DR is now described?

You mention manufacture metering software I'm not sure that is relevant. It is my understanding that on any given image the center tone, or Zone 5, should always equal the same 50% grey on all camera systems if using the same iso/aperture/shutter speed and standard processing. I believe the dynamic range has been traditionally considered as the number of stops higher/lower from the baseline exposure that a given sensor/negative can provide information for Zones 1 and 9.

I know that with digital photography the floor of an image is defined by noise while only the ceiling is defined by clipping. This is allows for the workflow that you and others have described of exposing for the highlights. I've used this technique myself thousands of times. I however, prefer not to have to underexpose my images like this. I prefer to not have to push Zone 5. Instead, if the scene allows for it, I'd rather be able to set Zone 5 right where I want the mid tones to be and have a sensor that will provide enough headroom for flexibility with where I process Zone 9 to be.

In any case it seems to me that the information on how many stops of headroom a sensor will provide would be very useful information. It seems to me that this should be part of a thorough dynamic range review.

 

[Mikehit]

It is my understanding that on any given image the center tone, or Zone 5, should always equal the same 50% grey on all camera systems if using the same iso/aperture/shutter speed and standard processing.

The problem is that there is no hard and fast definition of ISO which gives the manufacturer quite bit of leeway, to the extent that when the Olympus micro-fourthirds cameras took a leap in performance with the E-M5, there was a very strong suspicion that this was in part because theyplayed fast and loose with ISO definitions (their 3200 was closer to someone else's 1600)
How do you standardise 'zone 5'?
Some people argue that the standard 18% grey is the wrong tone to use as 50% tone. And even then data on the sensor is useless until it is presented on screen or paper as a picture and that introduces the variables of what is 'standard processing'? With whose software?

With cameras, like any tool, it is a case of knowing its limitations and this is where dynamic range is among the most useful - learn where the upper and lower limits of usability are and learn how to work within them.
With film, no-one to the best of my knowledge talked about 'headroom' but they did talk about how the film rendered highlights and shadows and how sudden the 'tail off of the sigmoid curve was. But digital sensors are linear response so that is irrelevant and it becomes a case of where you choose to put the '50%' tone. If you don't agree with the way the camera calculates it dial in a permanent '2/3 over exposure' or your preferred value (some people do).

 

[Aglet]

It is my understanding that on any given image the center tone, or Zone 5, should always equal the same 50% grey on all camera systems if using the same iso/aperture/shutter speed and standard processing.

The problem is that there is no hard and fast definition of ISO which gives the manufacturer quite bit of leeway, to the extent that when the Olympus micro-fourthirds cameras took a leap in performance with the E-M5, there was a very strong suspicion that this was in part because theyplayed fast and loose with ISO definitions (their 3200 was closer to someone else's 1600)
How do you standardise 'zone 5'?
Some people argue that the standard 18% grey is the wrong tone to use as 50% tone. And even then data on the sensor is useless until it is presented on screen or paper as a picture and that introduces the variables of what is 'standard processing'? With whose software?

With cameras, like any tool, it is a case of knowing its limitations and this is where dynamic range is among the most useful - learn where the upper and lower limits of usability are and learn how to work within them.
With film, no-one to the best of my knowledge talked about 'headroom' but they did talk about how the film rendered highlights and shadows and how sudden the 'tail off of the sigmoid curve was. But digital sensors are linear response so that is irrelevant and it becomes a case of where you choose to put the '50%' tone. If you don't agree with the way the camera calculates it dial in a permanent '2/3 over exposure' or your preferred value (some people do).

+1

If what you really want is your own flexible tone curve you can do that. But you first have to collect the data you want to work with and that means fitting the scene into the sensor's DR in such a way that you will get the data to process as you like later instead of relying on the mfr's metering and tone-curve. You have to know your equipments' limitations to do so.

This may also require you to do an extensive color calibration as well since the tone curves for R, G, and B may not be exactly the same all thru the tonal (zone) range, even in raw, which may result in some hue shifts when you make adjustments with a master (luminance?) tone curve. Otherwise you'll be tweaking individual RGB tone curves too.

This tends to be very pronounced with some mfrs when it comes to their in-camera jpg engines; certain jpg renderings can have significant hue shifts with slight exposure changes. Hence the variety of jpg renderings available like "faithful" and "neutral" which are handy if you're shooting fast action and need to post process jpgs.
I've learned to make use of this in some of the ML systems I use as the WYSIWYG EVF display shows me exactly the effect I get when I make exposure compensation adjustments on subjects that may have deeply saturated colors. This can save me from doing post work at all sometimes, as I can nail the effect I want completely in-camera at the time of shooting.

 

[Maiaibing]

It is striking how many of the "advances" are trending towards being able to "recover" a blown shot.

Coming soon, mind-reading AF,

Canon can do AF where your own eye decides the AF-point. A winning idea. Why they do not reintroduce this is beyond me.

That, and the overwhelmingly useful auto eye-AF of the competition would be more than nice.

 

[Sporgon]

Its interesting how the ability to recover highlights, or at least the useable highlight headroom is never tested or discussed.

Yes...that's because most discussions about DR are really about shadow pull using iso invariance. True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with and have a contrast scale beyond the legacy CCD capability of 8.4 stops. It's similar to slide film in it's DR, but way below the capabilities of negative print film's 14 stops of DR. The current fad with shadow recovery has little to do with DR. In the current Nikon and Sony sensors, yes you can pull shadows on a 100 iso image from apparent blackness to 5 stops of mid tones. But the equivalent iso noise is still 5 stops over the base iso 100. That's going to be iso 3200 noise in the shadows > new mid tones. That's really too high and it's not true HDR, it's just pulling noisy shadows using an iso trick. Some cameras do the same trick to make up their different iso values. Often 100 / 400/ 1600 iso values are native and all the other settings are based on one of these iso values and then the camera pulls the difference from pulling the exposure by a stop or two.

because, there's no such thing as "recoverable highlights"

Oh no ? What is clipped / blown with to day's processors may not be with tomorrows.

Go run some of your CR2 files through the ACR 2003 processes and compare 'blown' highlights with the latest process.

And regarding shadow lifting, lowlights at 0 lift as readily as highlights at 255 are pulled. (Forgot the <sarcasm> tag here

 

[Sabaki]

Its interesting how the ability to recover highlights, or at least the useable highlight headroom is never tested or discussed.

Yes...that's because most discussions about DR are really about shadow pull using iso invariance. True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with and have a contrast scale beyond the legacy CCD capability of 8.4 stops. It's similar to slide film in it's DR, but way below the capabilities of negative print film's 14 stops of DR. The current fad with shadow recovery has little to do with DR. In the current Nikon and Sony sensors, yes you can pull shadows on a 100 iso image from apparent blackness to 5 stops of mid tones. But the equivalent iso noise is still 5 stops over the base iso 100. That's going to be iso 3200 noise in the shadows > new mid tones. That's really too high and it's not true HDR, it's just pulling noisy shadows using an iso trick. Some cameras do the same trick to make up their different iso values. Often 100 / 400/ 1600 iso values are native and all the other settings are based on one of these iso values and then the camera pulls the difference from pulling the exposure by a stop or two.

because, there's no such thing as "recoverable highlights"

Oh no ? What is clipped / blown with to day's processors may not be with tomorrows.

Go run some of your CR2 files through the ACR 2003 processes and compare 'blown' highlights with the latest process.

And regarding shadow lifting, lowlights at 0 lift as readily as highlights at 255 are pulled.

Is HDR then the interim solution?

 

[hendrik-sg]

Its interesting how the ability to recover highlights, or at least the useable highlight headroom is never tested or discussed.

Yes...that's because most discussions about DR are really about shadow pull using iso invariance. True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with and have a contrast scale beyond the legacy CCD capability of 8.4 stops. It's similar to slide film in it's DR, but way below the capabilities of negative print film's 14 stops of DR. The current fad with shadow recovery has little to do with DR. In the current Nikon and Sony sensors, yes you can pull shadows on a 100 iso image from apparent blackness to 5 stops of mid tones. But the equivalent iso noise is still 5 stops over the base iso 100. That's going to be iso 3200 noise in the shadows > new mid tones. That's really too high and it's not true HDR, it's just pulling noisy shadows using an iso trick. Some cameras do the same trick to make up their different iso values. Often 100 / 400/ 1600 iso values are native and all the other settings are based on one of these iso values and then the camera pulls the difference from pulling the exposure by a stop or two.

Sure a 5 stop shadows push on a ISO 100 shot gives a ISO 3200 shot in the shadows. But a original ISO 3200 shot would blow the highlights, which a ISO 100 shot retains.

What the highlight recovery means, in my understanding if all color chanels are full, there is no more color information. How much it needs until the chanels are full should be defined by the base ISO, the lower the base ISO is, the more space for photons until the sensor overflows.

The dynamic range should be the quotient of "number of photons for overflowing" divided by "noise level in complete darkness".

 

[hendrik-sg]

Its interesting how the ability to recover highlights, or at least the useable highlight headroom is never tested or discussed.

Yes...that's because most discussions about DR are really about shadow pull using iso invariance. True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with and have a contrast scale beyond the legacy CCD capability of 8.4 stops. It's similar to slide film in it's DR, but way below the capabilities of negative print film's 14 stops of DR. The current fad with shadow recovery has little to do with DR. In the current Nikon and Sony sensors, yes you can pull shadows on a 100 iso image from apparent blackness to 5 stops of mid tones. But the equivalent iso noise is still 5 stops over the base iso 100. That's going to be iso 3200 noise in the shadows > new mid tones. That's really too high and it's not true HDR, it's just pulling noisy shadows using an iso trick. Some cameras do the same trick to make up their different iso values. Often 100 / 400/ 1600 iso values are native and all the other settings are based on one of these iso values and then the camera pulls the difference from pulling the exposure by a stop or two.

because, there's no such thing as "recoverable highlights"
either the tonal data is there or it's clipped. If it's clipped it's not recoverable.

At the shadows end the tonality is lost in noise and quantization steps if the noise is not too great. That's why reductions in read noise are a big deal when it comes to improving the DR.

There may be no such thing as "recoverable highlights" but it seems to me that there should be a way to test where the sensor clips the highlights. Since the noise is all in the shadows it seems like having a higher clip point for the brights would be extremely valuable.

For my work I'd much rather have the ability to expose for the mid tones where most of the information is, and then bring up/down the shadows/highlights as needed.

then expose lower, that the higlights are not blown and push the shadowes/mid tones

 

[3kramd5]

Canon can do AF where your own eye decides the AF-point. A winning idea. Why they do not reintroduce this is beyond me.

That, and the overwhelmingly useful auto eye-AF of the competition would be more than nice.

I have it on my Elan, but it never works quite well enough for my taste, and that's with only 7 points. With 61? Good luck!

 

[Loswr]

Canon can do AF where your own eye decides the AF-point. A winning idea. Why they do not reintroduce this is beyond me.

That, and the overwhelmingly useful auto eye-AF of the competition would be more than nice.

I have it on my Elan, but it never works quite well enough for my taste, and that's with only 7 points. With 61? Good luck!

Well, it would be better than iTR, which is an 'utter failure'. Or so Rishi claims.

 

[Sporgon]

Its interesting how the ability to recover highlights, or at least the useable highlight headroom is never tested or discussed.

Yes...that's because most discussions about DR are really about shadow pull using iso invariance. True high DR would allow us to recover highlights and shadows with the same levels of iso noise as the image was shot with and have a contrast scale beyond the legacy CCD capability of 8.4 stops. It's similar to slide film in it's DR, but way below the capabilities of negative print film's 14 stops of DR. The current fad with shadow recovery has little to do with DR. In the current Nikon and Sony sensors, yes you can pull shadows on a 100 iso image from apparent blackness to 5 stops of mid tones. But the equivalent iso noise is still 5 stops over the base iso 100. That's going to be iso 3200 noise in the shadows > new mid tones. That's really too high and it's not true HDR, it's just pulling noisy shadows using an iso trick. Some cameras do the same trick to make up their different iso values. Often 100 / 400/ 1600 iso values are native and all the other settings are based on one of these iso values and then the camera pulls the difference from pulling the exposure by a stop or two.

because, there's no such thing as "recoverable highlights"

Oh no ? What is clipped / blown with to day's processors may not be with tomorrows.

Go run some of your CR2 files through the ACR 2003 processes and compare 'blown' highlights with the latest process.

And regarding shadow lifting, lowlights at 0 lift as readily as highlights at 255 are pulled.

Is HDR then the interim solution?

A bracketed and blended shot will have better tonality at the bottom of the file compared with under exposing to gain more highlight room.

However HDR often isn't necessary if you don't under expose to start with. It is remarkable how much you can over expose these recent Canon cameras and hold good highlight detail if you use the latest ACR.

 

[Viggo]

"More moderate 3-4 ev pushes"

 

[dsnook]

It is my understanding that on any given image the center tone, or Zone 5, should always equal the same 50% grey on all camera systems if using the same iso/aperture/shutter speed and standard processing.

The problem is that there is no hard and fast definition of ISO

How do you standardise 'zone 5'?

what is 'standard processing'? With whose software?

I hadn't realized that this was not standardized. I thought that if you take an old spot meter it will give you the ISO/Shutter/Aperture that will make your subject 18% grey. This would then equal the 50% tone on any final image when using the mfr recommended processing. This is what I meant by standard processing, whether JPG output, default settings in ACR, or film development times. I had thought that the midpoint was supposed to be reasonably constant across platforms with the high/low clipping, noise, or tone curve varying.

Of course mfrs have always adjusted their processing to compensate for their ISO. They did this with film to allow for greater pushing latitude and from what you say it sounds like Olympus may have done the same thing recently. Of course this baseline processing is rarely how you would you would want to use film and it’s not how I shoot digitally either. Pushing and pulling is always happening, but I had always thought it was based around a standardized mid tone, which is why the same exposure settings will work across multiple camera systems.

With cameras, like any tool, it is a case of knowing its limitations and this is where dynamic range is among the most useful - learn where the upper and lower limits of usability are and learn how to work within them.

I completely agree with this statement. This is kinda part of my point and I think what Sporgon was getting at as well. I have learned the limitations of my own cameras and know when I need to underexpose (or more likely switch to bracketing/HDR) to get the final image I want. I don't have any problem with that. The issue is with cameras that I don't have but am considering. I would like to see reviews that provide information about the upper limits of the camera along with the lower.

If what you really want is your own flexible tone curve you can do that. But you first have to collect the data you want to work with and that means fitting the scene into the sensor's DR in such a way that you will get the data to process as you like later instead of relying on the mfr's metering and tone-curve. You have to know your equipments' limitations to do so.

Precisely, I want know the equipment’s limitations. I want to know if I shoot the same scene with my 5DmkIII and a 5DmkIV with the same iso/shutter/aperture where the highlights get clipped as well as where the noise floor is.