Who said Canon cameras suck?!?

[jrista]

D800. Its the name of the camera on everyones mouths these days. In all the blogs. In all the reviews. Its the thing raving Nikon fans rave about. Canon's taken a solid back seat to Nikon these days, particularly to the D800. I've said many times that Canon cameras have some pretty amazing highlight recovery, as Canon tends to tune their cameras to favor highlights (either intentfully or simply as a byproduct of their manufacturing process, I can't say...although I'm inclined to think its intentional given Canon popularity among wedding photographers.)

I regularly repeat that anecdote in many of my posts...but I just came across a couple accidental overexposures of some of my own photos that I think clearly demonstrate the point. While out photographing birds with a rental Canon EF 300mm f/2.8 L II IS and Canon EF 2x TC III, I kept coming across dragonflys. A telephoto lens with a TC is a great way to photograph some frame-filling insect "macros" (more like pseudomacro) without scaring the subject off. I accidentally set my exposure wrong and totally blew the first few shots:


1/100s @ f/5.6 ISO 100

The exposure should have been around 1/1000s @ f/8 ISO 100 (which I proved with some subsequent shots, which ended up being 1/1000s @ f/7.1 ISO 160)...so my exposure above was almost four stops overexposed. Thanks to the power of Lightroom 4.1 and its amazing highlight recovery, the above image, with -4 EV exposure correction and 60% highlight recovery, turned into this:



I'd experienced Canon's amazing highlight headroom when photographing the moon. I REALLY push my moon exposures...to the point where once exposed the moon looks like a nearly uniform almost-white disc in the in-camera preview. Once imported, its clear that there are actually few parts of the moon that are actually white. I'd never actually overexposed something so much that on import it really DID look almost entirely white. The histogram of the dragonfly was all bunched up in two peaks near the very far right...with a small gap between the second peak and the actual right edge...a gap maybe 1 or 2 pixels wide. With 100% highlight recovery in LR 4.1, even the specular highlights on the wings still retain a lot of detail:



Since this image started out way overexposed, there is zero color or pattern noise in the shadows. There is also minimal random (photon shot) noise in the shadows as well...they look as clean as a D800 at ISO 100! ;-)

So, the next time someone tells you Canon cameras suck...send 'em here. While Canon cameras may not be able to achieve 13.2 stops of DR or allow noiseless shadow recovery like the D800 can, they really do know how to pack in the highlights, and maintain full color fidelity while recovering. The next time you need low noise shadows...expose to the right....then, try exposing farther to the right. 

[poias]

Atleast you admit that having that ABILITY is nice, unlike those who claim that sensor does not matter.

Now imagine having all 14.4 stops of D800 DR rather than 11 DR that 5D3 has.

[jrista]

Atleast you admit that having that ABILITY is nice, unlike those who claim that sensor does not matter.

Now imagine having all 14.4 stops of D800 DR rather than 11 DR that 5D3 has.

I really don't know how to put this one to rest. The D800 does NOT have 14.4 stops of dynamic range. Not as a physical capability. That is a fabrication, thanks to the clever way DXO's software DOWNSAMPLES images. So, a few things here. First off...strait out of the camera, even DXO claims the D800 has 13.2 stops of DR. That is hardware dynamic range...we'll get back to that. Second, DXO also clearly claims that to acheive the "wonderous 14.4 stops of DR", they had to downscale that HUGE 36.3mp D800 image down to...a measly 8mp 8x12" print. If your buying a D800 to print at 8x12", your wasting a hell of a lot of money. Second, most people simply use some form of basic averaging to scale their images, which is not going to result in 1.2 stops of additional shadow DR...not by a long shot. You AVERAGE the pixels in the image...including shadow pixels.

For those who don't actually downscale their images, or who crop, your only going to get what the hardware is capable of. DXO measures that at 13.2 stops. I don't much care for DXO, but I still trust that within the microuniverse that is DXO, their results are consistent and "accuratE". If DXO ever came out and claimed that the unmodified RAW was capable of 14.0 stops or more, I would immediately lose the last ounce of trust I have in them, since its impossible to capture more stops of DR than the bit depth of the sensor/adc/image processor. At best you could probably extract...assuming the best real world efficiency possible, 13.9 stops of DR, possibly to another place or two of precision. So long as the final DIGITAL output is still 14 bits, you can't do any better than that. If you intend to use all the pixels the D800 has to offer and NOT downscale, then you can't get 14.4 stops of DR, regardless of how clever your downscaling algorithm may be. Your getting the hardware's 13.2 stops of native DR (what DXO calls Screen DR).

A lot of people will argue that you can use cleaver dithering algorithms when downscaling to expand the dynamic range in the shadows. They are absolutely correct, you can...and DXO's specialized software probably does. But for 99% of people who use normal-people tools, such as Photoshop, to do their scaling...they are using decades old algorithms like bilinear or bicubic scaling. Basic filters based on your simple averaging algorithm. Shadows are averaged, not dithered and deepened, when scaling...so while most people might gain a small amount of DR improvement when downscaling due to noise averaging, they aren't going to be significantly deepening the shadows...and certainly not by 2.4x (1.2 stops).

Finally, its been demonstrated by a number of reviewers around the net that Nikon tends to overexpose a bit relative to Canon cameras. I suspect thats due to the fact that they allocate more levels to the shadows (considerably more than most other camera manufacturers), which is a significant reason why they have such amazing shadow recovery...its not solely due to lower read noise (which would restore a few more levels of luminance to full usability). That leaves fewer levels for the highlights, which is often why you see blown or very nearly blown highlights in sample photos and videos comparing Nikon cameras to Canon cameras in extreme DR scenarios.

I'm not arguing that Canon can achieve 13.2 stops of DR like a Sony Exmor sensor can, but I believe they are capable of more DR than they are generally given credit for...due to how they allocate levels during read. I was able to massively over-expose the photo in my original post, such that most of it looked white, and recover. A D800 wouldn't be able to do the same...and in all probability a 4-stop over-exposure would result in unrecoverable blown highlights. Now, on the same note, you wouldn't need to bother with ETTR, or not nearly as much, to recover shadows with the D800, and their low read noise and greater shadow level allocation still gives them the edge. But overexposures DO happen...they tend to be accidental, but they also have the tendency to be more damaging than an underexposure. You can almost always recover an underexposed image to a reasonable, even usable level...since you haven't actually LOST any information. If you blow past maximum saturation in a digital sensor, those pixels are gone for good, and no amount of recovery will save your ass then. In those cases, I'll keep my Canon, thanks.

[LetTheRightLensIn]

There is no such thing as more highlight recovery. These designs are linear. Maybe they decide to call middle gray something a little lower and use a different default tone curve, but that is nothing to do with the sensor. And if you measure max well capacity and then shadow noise the story has already been told. You could shift middle gray for nikon and get the same highlight savings (assuming it's actually true they are placed differently in RAW suggestions) and still better shadows.

[LetTheRightLensIn]

Atleast you admit that having that ABILITY is nice, unlike those who claim that sensor does not matter.

Now imagine having all 14.4 stops of D800 DR rather than 11 DR that 5D3 has.

I really don't know how to put this one to rest. The D800 does NOT have 14.4 stops of dynamic range. Not as a physical capability. That is a fabrication, thanks to the clever way DXO's software DOWNSAMPLES images. So, a few things here. First off...strait out of the camera, even DXO claims the D800 has 13.2 stops of DR. That is hardware dynamic range...we'll get back to that. Second, DXO also clearly claims that to acheive the "wonderous 14.4 stops of DR", they had to downscale that HUGE 36.3mp D800 image down to...a measly 8mp 8x12" print. If your buying a D800 to print at 8x12", your wasting a hell of a lot of money. Second, most people simply use some form of basic averaging to scale their images, which is not going to result in 1.2 stops of additional shadow DR...not by a long shot. You AVERAGE the pixels in the image...including shadow pixels.

For those who don't actually downscale their images, or who crop, your only going to get what the hardware is capable of. DXO measures that at 13.2 stops. I don't much care for DXO, but I still trust that within the microuniverse that is DXO, their results are consistent and "accuratE". If DXO ever came out and claimed that the unmodified RAW was capable of 14.0 stops or more, I would immediately lose the last ounce of trust I have in them, since its impossible to capture more stops of DR than the bit depth of the sensor/adc/image processor. At best you could probably extract...assuming the best real world efficiency possible, 13.9 stops of DR, possibly to another place or two of precision. So long as the final DIGITAL output is still 14 bits, you can't do any better than that. If you intend to use all the pixels the D800 has to offer and NOT downscale, then you can't get 14.4 stops of DR, regardless of how clever your downscaling algorithm may be. Your getting the hardware's 13.2 stops of native DR (what DXO calls Screen DR).

A lot of people will argue that you can use cleaver dithering algorithms when downscaling to expand the dynamic range in the shadows. They are absolutely correct, you can...and DXO's specialized software probably does. But for 99% of people who use normal-people tools, such as Photoshop, to do their scaling...they are using decades old algorithms like bilinear or bicubic scaling. Basic filters based on your simple averaging algorithm. Shadows are averaged, not dithered and deepened, when scaling...so while most people might gain a small amount of DR improvement when downscaling due to noise averaging, they aren't going to be significantly deepening the shadows...and certainly not by 2.4x (1.2 stops).

Finally, its been demonstrated by a number of reviewers around the net that Nikon tends to overexpose a bit relative to Canon cameras. I suspect thats due to the fact that they allocate more levels to the shadows (considerably more than most other camera manufacturers), which is a significant reason why they have such amazing shadow recovery...its not solely due to lower read noise (which would restore a few more levels of luminance to full usability). That leaves fewer levels for the highlights, which is often why you see blown or very nearly blown highlights in sample photos and videos comparing Nikon cameras to Canon cameras in extreme DR scenarios.

I'm not arguing that Canon can achieve 13.2 stops of DR like a Sony Exmor sensor can, but I believe they are capable of more DR than they are generally given credit for...due to how they allocate levels during read. I was able to massively over-expose the photo in my original post, such that most of it looked white, and recover. A D800 wouldn't be able to do the same...and in all probability a 4-stop over-exposure would result in unrecoverable blown highlights. Now, on the same note, you wouldn't need to bother with ETTR, or not nearly as much, to recover shadows with the D800, and their low read noise and greater shadow level allocation still gives them the edge. But overexposures DO happen...they tend to be accidental, but they also have the tendency to be more damaging than an underexposure. You can almost always recover an underexposed image to a reasonable, even usable level...since you haven't actually LOST any information. If you blow past maximum saturation in a digital sensor, those pixels are gone for good, and no amount of recovery will save your ass then. In those cases, I'll keep my Canon, thanks.

So don't nornalize to 8MP, normalized to 36MP and you'll still see the DR advantage that large.

[jrista]

Atleast you admit that having that ABILITY is nice, unlike those who claim that sensor does not matter.

Now imagine having all 14.4 stops of D800 DR rather than 11 DR that 5D3 has.

I really don't know how to put this one to rest. The D800 does NOT have 14.4 stops of dynamic range. Not as a physical capability. That is a fabrication, thanks to the clever way DXO's software DOWNSAMPLES images. So, a few things here. First off...strait out of the camera, even DXO claims the D800 has 13.2 stops of DR. That is hardware dynamic range...we'll get back to that. Second, DXO also clearly claims that to acheive the "wonderous 14.4 stops of DR", they had to downscale that HUGE 36.3mp D800 image down to...a measly 8mp 8x12" print. If your buying a D800 to print at 8x12", your wasting a hell of a lot of money. Second, most people simply use some form of basic averaging to scale their images, which is not going to result in 1.2 stops of additional shadow DR...not by a long shot. You AVERAGE the pixels in the image...including shadow pixels.

For those who don't actually downscale their images, or who crop, your only going to get what the hardware is capable of. DXO measures that at 13.2 stops. I don't much care for DXO, but I still trust that within the microuniverse that is DXO, their results are consistent and "accuratE". If DXO ever came out and claimed that the unmodified RAW was capable of 14.0 stops or more, I would immediately lose the last ounce of trust I have in them, since its impossible to capture more stops of DR than the bit depth of the sensor/adc/image processor. At best you could probably extract...assuming the best real world efficiency possible, 13.9 stops of DR, possibly to another place or two of precision. So long as the final DIGITAL output is still 14 bits, you can't do any better than that. If you intend to use all the pixels the D800 has to offer and NOT downscale, then you can't get 14.4 stops of DR, regardless of how clever your downscaling algorithm may be. Your getting the hardware's 13.2 stops of native DR (what DXO calls Screen DR).

A lot of people will argue that you can use cleaver dithering algorithms when downscaling to expand the dynamic range in the shadows. They are absolutely correct, you can...and DXO's specialized software probably does. But for 99% of people who use normal-people tools, such as Photoshop, to do their scaling...they are using decades old algorithms like bilinear or bicubic scaling. Basic filters based on your simple averaging algorithm. Shadows are averaged, not dithered and deepened, when scaling...so while most people might gain a small amount of DR improvement when downscaling due to noise averaging, they aren't going to be significantly deepening the shadows...and certainly not by 2.4x (1.2 stops).

Finally, its been demonstrated by a number of reviewers around the net that Nikon tends to overexpose a bit relative to Canon cameras. I suspect thats due to the fact that they allocate more levels to the shadows (considerably more than most other camera manufacturers), which is a significant reason why they have such amazing shadow recovery...its not solely due to lower read noise (which would restore a few more levels of luminance to full usability). That leaves fewer levels for the highlights, which is often why you see blown or very nearly blown highlights in sample photos and videos comparing Nikon cameras to Canon cameras in extreme DR scenarios.

I'm not arguing that Canon can achieve 13.2 stops of DR like a Sony Exmor sensor can, but I believe they are capable of more DR than they are generally given credit for...due to how they allocate levels during read. I was able to massively over-expose the photo in my original post, such that most of it looked white, and recover. A D800 wouldn't be able to do the same...and in all probability a 4-stop over-exposure would result in unrecoverable blown highlights. Now, on the same note, you wouldn't need to bother with ETTR, or not nearly as much, to recover shadows with the D800, and their low read noise and greater shadow level allocation still gives them the edge. But overexposures DO happen...they tend to be accidental, but they also have the tendency to be more damaging than an underexposure. You can almost always recover an underexposed image to a reasonable, even usable level...since you haven't actually LOST any information. If you blow past maximum saturation in a digital sensor, those pixels are gone for good, and no amount of recovery will save your ass then. In those cases, I'll keep my Canon, thanks.

So don't nornalize to 8MP, normalized to 36MP and you'll still see the DR advantage that large.

Yes, you will...you'll see a shadow DR advantage. But it will be 13.2 STOPS, NOT 14.4!!! Thats a difference of 1.2 stops, or a 2.4x light difference...a HUGE difference, a meaningful difference! From a real-world perspective, for people who use the D800 for what it is, a 36.3mp megapixel monster, no one is going to get 14.4 stops of DR out of the D800. In the real world, no one is actually doing "comparisons" between the 5D III, 1D X, or any other camera and the D800...they are using the damnable thing to take photographs and publish them (most probably in print.)

Lets stop talking about the D800 as it its simply and solely a trophy to be compared, and start talking about it from a real world context. No one cares how it compares if you upscale a 5D III image to 36.3mp size. Neither does anyone who uses a 5D III or any other Canon camera really care how it compares if you downscale a D800 image to 22.3mp size. They care whats possible in the real world, with real-world software...tone curves and all.

[jrista]

There is no such thing as more highlight recovery. These designs are linear. Maybe they decide to call middle gray something a little lower and use a different default tone curve, but that is nothing to do with the sensor. And if you measure max well capacity and then shadow noise the story has already been told. You could shift middle gray for nikon and get the same highlight savings (assuming it's actually true they are placed differently in RAW suggestions) and still better shadows.

If you open up a RAW in a DCRAW program without applying any kind of tone curve at all, yes, things are linear. But in the real world, people don't open their photos in a basic DCRAW editor without a tone curve. All other real-world RAW editing software applies a tone curve. Nikon tone curves have a very LONG foot at the shadow end of their curves, relative to Canon's. Canon tone curves have a slightly longer shoulder in the highlights than Nikon tone curves. I'm not denying that Exmor's low read noise is a benefit...it most certainly is, and you can utilize more levels in the shadows than any other sensor on earth with it...but its not the entire story with the D800's "shadow recovery" (an aspect of editing that wouldn't even exist if everyone opened their RAW images in an editor without any kind of base tone curve applied...you would start with maximum DR and push down the shadows from there, rather than the other way around).

[PVS]

Why people mistake DR for exposure latitude?

[jrista]

Why people mistake DR for exposure latitude?

Technically speaking, I am talking more about exposure latitude than DR...what you can do with a RAW image once it has been imported into a RAW editor for post-processing. The fact that a tone curve is applied that affects the observed exposure is the key factor I'm referring to. With a Canon camera, thanks to their tone curves, when they display a photo on the LCD, even if the highlights appear blown and the image completely over exposed, there is still PLENTY of room (apparently, nearly 4 stops worth) to "recover".

The D800, D7000, probably even the D600, all have more DR, but its allocated to the shadows by default (thanks to that long foot on the shadow end of their default tone curves.) If you look at a photograph on the the D800 (or D7000, which is actually what I have direct experience with) and the highlights appear blown and the image completely over exposed, the chances are you've actually literally blown some of your highlights to the point where they are not recoverable (you might have a stop or two worth of "recovery" room with Nikon's default tone curves.) So long as you generally underexpose with a Nikon camera that has a Sony Exmor sensor, and lift shadows in post, you'll never have a problem...and thats the true benefit of Nikon. But you have to actually make sure you never overexpose.

[LetTheRightLensIn]

Lets stop talking about the D800 as it its simply and solely a trophy to be compared, and start talking about it from a real world context. No one cares how it compares if you upscale a 5D III image to 36.3mp size. Neither does anyone who uses a 5D III or any other Canon camera really care how it compares if you downscale a D800 image to 22.3mp size. They care whats possible in the real world, with real-world software...tone curves and all.

real world it's simply not fair to compare cameras on a non-normalized basis to one another,it doesn't make any sense

[Bennymiata]

Some of you Nikon D800 fanboys really make me sick.
You think the D800 is the be-all and end-all of all DSLR cameras, when in fact it has some glaring faults like poor autofocus (especially on the left side) and a rear screen that makes everything look green.

I do a lot of commercial shots where colour is very important, and if I had to go by the rear screen, a D800 would drive me bonkers!
I can and do go by the screen on my 5D3 and it is very accurate, unlike the Nikon screen.

Put properly exposed shots of the D800 beside shots from a 5D3 on a good quality computer screen and you would have trouble picking the differences, except the Canon's colours are more true to life.

Even if the D800 had a thousand megapickles, the 5D3 is still a better all-around camera and is certainly my tool of choice for the jobs I do.
In fact, with all the types of photography I do for a living, or for my own fun, it has always done a sterling job, even in very difficult and demanding situations.

You guys who carry on about how much better a D800 is than a 5D3 remind me of the immature little boys who say that their car is better than yours, because it can do 0-60 1/10th of a second quicker, yet it rides like a buckboard and handles like a limp rag.
There is far more to a good camera than a heap of megapickles, just as there is a lot more to a good car than a quick 0-60 time.

[NormanBates]

Why people mistake DR for exposure latitude?

I always thought they were the same. Care to explain the difference?

[verysimplejason]

Why people mistake DR for exposure latitude?

I always thought they were the same. Care to explain the difference?

Exposure latitude is the extent to which a light-sensitive material can be overexposed or underexposed and still achieve an acceptable result. Since the acceptability of the result is dependent on both personal aesthetics and artistic intentions, the measurement of exposure latitude is, by definition, somewhat subjective. However, the relative differences between mediums are generally agreed upon: reversal film tends to have very little latitude, color negative film has considerably more, and digital sensors slot between the two.
It is not to be confused with dynamic range, the range of light intensities a medium can capture simultaneously. A recording medium with greater dynamic range will be able to record more details in the dark and light areas of a picture. Latitude depends on dynamic range. If the same scene can be recorded using less than the full brightness range available to the medium, the exposure can be shifted along the range without losing information in the shadows or highlights. Greater exposure latitude allows one to compensate for errors in exposure while retaining quality.
Professional critique of digital cine cameras often centers on the extent to which their dynamic range, and exposure latitude by extension, falls short of that of negative film.

This is from Wiki.

http://en.wikipedia.org/wiki/Exposure_latitude

[LetTheRightLensIn]

Some of you Nikon D800 fanboys really make me sick.
You think the D800 is the be-all and end-all of all DSLR cameras, when in fact it has some glaring faults like poor autofocus (especially on the left side) and a rear screen that makes everything look green.

I do a lot of commercial shots where colour is very important, and if I had to go by the rear screen, a D800 would drive me bonkers!
I can and do go by the screen on my 5D3 and it is very accurate, unlike the Nikon screen.

Put properly exposed shots of the D800 beside shots from a 5D3 on a good quality computer screen and you would have trouble picking the differences, except the Canon's colours are more true to life.

Even if the D800 had a thousand megapickles, the 5D3 is still a better all-around camera and is certainly my tool of choice for the jobs I do.
In fact, with all the types of photography I do for a living, or for my own fun, it has always done a sterling job, even in very difficult and demanding situations.

You guys who carry on about how much better a D800 is than a 5D3 remind me of the immature little boys who say that their car is better than yours, because it can do 0-60 1/10th of a second quicker, yet it rides like a buckboard and handles like a limp rag.
There is far more to a good camera than a heap of megapickles, just as there is a lot more to a good car than a quick 0-60 time.

we are talking sensor only here nothing more

other than low ISO DR the 5D3 really is quite awesome, 6fps FF, 1 series AF, compact body size, now with ML the video is quite usable and soon it should offer better compression and be a really nice video solution, nice UI (more MP would be nice but you can't have it all yet perhaps so it's really just the low ISO DR that was the one unfortunate thing, other great)

[verysimplejason]

It's the photographer that sucks and not the camera most of the time.    Modern DSLRs are basically good enough to capture pictures almost all the time.  I admit there are limitations but sometimes it's the skill that's limited and not the camera.