Dynamic Range - Try it for yourself, conclude for yourself: 5D III vs. A7r

[Marsu42]

The default conversion gives you much less dynamic range than is available in the raw data. Even if both highlights and shadows are clipped in the default conversion, using -highlights and +shadows may recover several stops of DR from the ends of the histogram.

Fyi: The effect is far stronger on crop cameras (at least my 60d) vs. full frame (on my 6d). I guess for the crop cameras, Canon adds a larger "safety limit" to prevent clipped highlights when shooting raw, but expects "pros" on ff to know what they're doing?

When you shoot RAW, the camera generates a small JPG preview image that's embedded in the RAW file container. That image is what you see during the on-camera review, and it's what's used to generate the histogram and highlight warning (blinkies). All in-camera settings are applied (color space, ALO, picture style, HTP, etc.).

The nice thing about this is that you don't waste card space with a jpeg sidecar, but can extract it from the cr2 anytime later. The thumbnail is also very handy for quick browsing/rating (except in ACR), so getting the jpeg settings right makes sense even when shooting raw.

If you don't mind somewhat funky images for on-camera review, you can use UniWB to get a histogram that better approximates the RAW data.

The length people w/o ML have to go to to get a raw histogram approximation ...

 

[unfocused]

After 25 pages this thread is starting to provide some useful information. How disappointing.

Can we go back to lifting shadows by five stops and arguing over banding?

 

[heptagon]

After 25 pages this thread is starting to provide some useful information. How disappointing.

Can we go back to lifting shadows by five stops and arguing over banding?

With the 6D, the banding issues are basically gone. 5DIII shooters just use the wrong tool for the job and should get add a 6D to their kit.

 

[Lee Jay]

Read noise aside, larger sensors have more dynamic range.

Yeah, but that's misleading, because read noise scales (down) along with size (well capacity), so DR isn't strongly affected by pixel size.

 

[Lee Jay]

It looks like the sensorgen.info domain has expired, so I can't provide any references.

https://web.archive.org/web/20140708054926/http://sensorgen.info/

 

[AcutancePhotography]

After 25 pages this thread is starting to provide some useful information. How disappointing.

Can we go back to lifting shadows by five stops and arguing over banding?

Weird huh? I had to check the URL to see if I accidently went to a real photography site. LoL

 

[Sporgon]

After 25 pages this thread is starting to provide some useful information. How disappointing.

Can we go back to lifting shadows by five stops and arguing over banding?

With the 6D, the banding issues are basically gone. 5DIII shooters just use the wrong tool for the job and should get add a 6D to their kit.

Don't you believe it, some of the acclaimed band masters here could manage it.

Back up a second. So even if you shoot RAW the histogram in camera is only displaying the color space you've chosen? So does that mean I should set my camera's color space to Adobe RGB just so I get a more accurate histogram?

When you shoot RAW, the camera generates a small JPG preview image that's embedded in the RAW file container. That image is what you see during the on-camera review, and it's what's used to generate the histogram and highlight warning (blinkies). All in-camera settings are applied (color space, ALO, picture style, HTP, etc.).

If you don't mind somewhat funky images for on-camera review, you can use UniWB to get a histogram that better approximates the RAW data.

Another interesting point to add to PBD's sRGB / Adobe RGB histogram info.

If I'm not wanting an OOC jpeg and just shooting for the raw file I set either neutral or faithful in Picture Styles, with contrast set to zero. This seems to give a reasonable impression of the raw data. Haven't tried UniWB but will do to see if that is better.

 

[DominoDude]

Something here that makes any sense? --> http://www.stanfordcomputeroptics.com/technology/dynamic-range/photon-noise.html

 

[Marsu42]

After 25 pages this thread is starting to provide some useful information. How disappointing.

Proves even people bound to argue some point over and over get tired at some point

This is not the case for Canon cameras because they have higher read noise, which eats away at a significant amount (at least two stops) of the maximum potential dynamic range allowed by a given sensor size and design/technology.

... but with current Canon sensors, you can argue that shooting on iso400 is nearly as good as iso100 - take that, Nikon trolls :-> ... so using the "min. iso" setting of auto iso really makes sense to get higher shutter speeds at (nearly) no iq cost.

 

[Lee Jay]

It looks like the sensorgen.info domain has expired, so I can't provide any references.

https://web.archive.org/web/20140708054926/http://sensorgen.info/

Case in point:

7D: 8.3e- (4.3µm)
70D: 13.5e- (4.11µm)

6D: 26.8e- (6.6µm)
5DIII: 33.1e- (6.25µm)

Cameras with smaller pixels sometimes have more read noise. Read noise doesn't really have that much to do with pixel size...it has to do with downstream electronics, like the ADC.

https://web.archive.org/web/20140715201224/http://www.sensorgen.info/CanonPowershot_G15.html

The reason the read noise scales with pixel size is that well capacity does. If you need more range to capture a larger voltage, you get more noise at the smaller voltages.

 

[Marsu42]

ISO 400 is still worse overall...the photon shot noise is higher because a lesser signal is amplified more, which impacts the entire signal. Read noise drops, which only affects the low signal (shadow) areas. I'd still choose ISO 100 if I want the lowest photon shot noise possible and the cleanest midtone through highlight detail.

Hmmyes, I know about the effect of read noise on low iso dr. But except for tripod shooting, I have discovered that my overall iq is much better if I use 200 or even 400 to gain more stopping power with higher shutter speed = less residual motion blur. But this is really just my personal, subjective impression.

Are there any sources or samples how much the difference of iso 100/200/400 on the latest Canon sensors is (like my 6d)? I don't worry about photon noise on ff lower iso, but I wonder how much other iq parts suffer (colors, tonality).

 

[Zv]

Back up a second. So even if you shoot RAW the histogram in camera is only displaying the color space you've chosen? So does that mean I should set my camera's color space to Adobe RGB just so I get a more accurate histogram?

When you shoot RAW, the camera generates a small JPG preview image that's embedded in the RAW file container. That image is what you see during the on-camera review, and it's what's used to generate the histogram and highlight warning (blinkies). All in-camera settings are applied (color space, ALO, picture style, HTP, etc.).

If you don't mind somewhat funky images for on-camera review, you can use UniWB to get a histogram that better approximates the RAW data.

Thanks, I had a brief flirt with video once and became familiar with picture style settings etc and ended on faithful with a low contrast setting. I've also had HTP and ALO off from day one. I think just knowing now that the histogram is for the JPEG preview is enough for me at this stage. I can work with that. Moving forward I think ML will be better.

 

[privatebydesign]

The histogram I referred to was LR, looking at the Raw files. If that is incorrect also, I'd appreciate a clarification.

Lightroom is a colour space marvel and I don't know the answer to every combination but as I understand it:

• The Library Module uses Adobe RGB for it's previews and thumbnails unless the file is tagged with a different colour space, normally sRGB, remember RAW files don't have a clour space attached so are displayed in Adobe RGB in the Library Module.
• The Develop Module displays the images in a custom Lightroom colour space commonly called Melissa RGB (named after Melissa Gaul an Adobe manager), this has the cromacity, or colours, of the Prophoto colour space but it has an sRGB gamma curve applied to it. Melissa RGB is what the histogram in the develop module is based on.
• I do not know 100% if the histogram in the Library Module is based on the Library Module's Adobe RGB preview or the behind the scenes Melissa RGB render, but I believe it is based on the behind the scenes Melissa RGB render.
• Under the hood of the Develop Module Lightroom actually works on the RAW files in another custom space, this has the Prophoto cromacity and a gamma value of 1. We never see images in this space but for internal calculations, apparently, it is simpler to do that on a gamma 1 file.
• The differences between Adobe RGB and Melissa RGB is the reason you get a colour shift on the same image between the Library Module and the Develop Module with RAW files, you shouldn't with tagged jpegs.
  

 

[LetTheRightLensIn]

But how do you control light out in the wild, in a forest or overlooking a ton of land?
Plus, in some cases, even if you could control it, you might be controlling away the magic.

It sounds like you are talking about planned type shots?

I think this suggestion is a perfect way to find out if the kinds of shooting situations a person finds themselves in requires more dynamic range. I put requires in italics because, if a shooter thinks carefully about what they're doing, I'll bet they can find a way to manage extreme contrast ranges in fast moving situations where HDR might not be possible.

I seldom encounter any problem. The reason is likely tied to my long history of controlling contrast from Back In The Day when the only light sensitive material available was film. Working in silver halide for far too many decades taught me the value of light controls to manage contrast.

... I suggest they go back and study the histograms of some of their more contrasty images and see how many (if any) would benefit from more DR.

 

[LetTheRightLensIn]

Back up a second. So even if you shoot RAW the histogram in camera is only displaying the color space you've chosen? So does that mean I should set my camera's color space to Adobe RGB just so I get a more accurate histogram?

Yes, and there are ways to set it to make it even closer to what the camera sees, although it makes the previews look really nasty. ML RAW Histogram is better.

One side note, you often see it claimed that digital sensors are terrible for red and clip red and Canon sensors are super bad at clipping red. But the reality is that it is sRGB that is clipping the reds, not the digital sensors. Canon sensors capture way the heck more red than sRGB. Even a simple red rose blows out sRGB. A lot of flowers always look weird and not like in real life, but it is sRGB not the Canon sensor that is the culprit. Keep the shot in ProphotoRGB 16bit and then view on a wide gamut monitor and you'll see for some stuff that a lot more color is there. It was just that sRGB clipped it.

I have also noticed the default conversion in LR seems to clip earlier than some of the others like Faithful or Neutral. I feel like I should be tinkering with that in some way to maximize latitude. For most of my stuff the highlight slider takes care of my DR needs. It's hard to gauge just how far to ettr without ML. I had it on the M but it got buggy with the EF-M 11-22mm lens for some reason so now I'm just doing without. Maybe I should install it on the 5D2. Not sure I need all those features but this RAW histogram thing has got me re-interested. ???

Yes.

One tricky thing is that absolute extreme ETTR can make processing tricky as most standard tone curves end leaving you with poor highlight separation and a lot of stock color profiles are twisted so you can get weird color tints and problems.

 

[LetTheRightLensIn]

Back up a second. So even if you shoot RAW the histogram in camera is only displaying the color space you've chosen?

And the in-camera processing (especially contrast and saturation) you've chosen.

Yeah dialing contrast way down and saturation down a little bit is a good idea too, if you want the regular histogram to give a better picture of what the RAW is doing.

 

[LetTheRightLensIn]

If anything, I've learned that, despite trying to ETTR by looking at the histogram on the camera, I can overexpose even more than I imagined. I knew that the histogram was only based on the jpg-thumbnails, just as the image I see on the display, and that it is a result of the PictureStyle and Colourspace.

The use of ML's Auto-ETTR and the RAW-based histograms have shown that I can be a lot more brutal when I choose my exposure. But the limiting parameter are the widest aperture of the chosen lens, and it dictates the combinations of shutter times and ISO I can pick to get my desired/needed exposure. I've learned that I, most of the time, need rather short shutter times to not introduce shake, and then I end up needing to increase the ISO to get my histogram to crawl over more to the right.

I think we could all benefit from a redesign of the Canon sensors. A redesign that incorporates deeper photon wells (more electrons can be stored and thus giving us a greater dynamic range), and measures taken to lessen the banding tendency. From texts I've read, by Stanford university researchers, the A/D-converters are less crucial for a high DR in optical sensors.

Yeah one nice thing about having more DR on the system is that it can also let you manage to get away with hand holdable exposures at lowest ISO or two more easily. And that is also a problem with using brackets and software to align, even if the alignment actually does work out well and not leave minor or mega defects, the brightest bracket might pick up handshake if you are in marginal conditions, while the highlight saving single shot and then lifting shadows from that might be easily hand-holdable.

 

[DominoDude]

If anything, I've learned that, despite trying to ETTR by looking at the histogram on the camera, I can overexpose even more than I imagined. I knew that the histogram was only based on the jpg-thumbnails, just as the image I see on the display, and that it is a result of the PictureStyle and Colourspace.

The use of ML's Auto-ETTR and the RAW-based histograms have shown that I can be a lot more brutal when I choose my exposure. But the limiting parameter are the widest aperture of the chosen lens, and it dictates the combinations of shutter times and ISO I can pick to get my desired/needed exposure. I've learned that I, most of the time, need rather short shutter times to not introduce shake, and then I end up needing to increase the ISO to get my histogram to crawl over more to the right.

I think we could all benefit from a redesign of the Canon sensors. A redesign that incorporates deeper photon wells (more electrons can be stored and thus giving us a greater dynamic range), and measures taken to lessen the banding tendency. From texts I've read, by Stanford university researchers, the A/D-converters are less crucial for a high DR in optical sensors.

Yeah one nice thing about having more DR on the system is that it can also let you manage to get away with hand holdable exposures at lowest ISO or two more easily. And that is also a problem with using brackets and software to align, even if the alignment actually does work out well and not leave minor or mega defects, the brightest bracket might pick up handshake if you are in marginal conditions, while the highlight saving single shot and then lifting shadows from that might be easily hand-holdable.

Precisely! You get more room to manoeuvre in. Less restrictions on which combinations to select from to get what you want and need. So it is, indeed, giving more than just technically better rendition on the pixel level, you get that extra artistic freedom to combine settings that's usable.

 

[deleteme]

If anything, I've learned that, despite trying to ETTR by looking at the histogram on the camera, I can overexpose even more than I imagined. I knew that the histogram was only based on the jpg-thumbnails, just as the image I see on the display, and that it is a result of the PictureStyle and Colourspace.

The use of ML's Auto-ETTR and the RAW-based histograms have shown that I can be a lot more brutal when I choose my exposure. But the limiting parameter are the widest aperture of the chosen lens, and it dictates the combinations of shutter times and ISO I can pick to get my desired/needed exposure. I've learned that I, most of the time, need rather short shutter times to not introduce shake, and then I end up needing to increase the ISO to get my histogram to crawl over more to the right.

I think we could all benefit from a redesign of the Canon sensors. A redesign that incorporates deeper photon wells (more electrons can be stored and thus giving us a greater dynamic range), and measures taken to lessen the banding tendency. From texts I've read, by Stanford university researchers, the A/D-converters are less crucial for a high DR in optical sensors.

Actually what that tells me is that I want an EVF with a live histogram so that I can optimize before I take the shot.
Virtually all final images will have data that is eliminated on purpose to create the final image. Optimizing capture is what we always did with film.
Yes, we fiddled in the darkroom but if we knew we wanted shadow detail we exposed for shadows, if we wanted highlights we exposed of them. When using transparency film this meant we bracketed.

 

[sarangiman]

The reason ISO 100 on a Canon has roughly the same DR as ISO 400 is because the read noise drops by about a stop with each progressive increase to ISO. Hence the flattening of the DR curve on Canon sensors.

Well, it's not that the read noise drops, it's just that all other data & noise off the sensor is amplified 4x compared to ISO 100, which lowers the impact of the read noise. So even though you're throwing away 2 stops of highlight range at ISO 400, 4x (2 EV) darker signal is now being amplified to be brought up to your SNR = 1 (or what have you) threshold, leaving DR largely the same. Cameras with very little downstream read noise don't need the sensor signal amplified to overcome the 30+ electrons of read noise in a Canon camera.

It is still Exmor...I just think they removed the black point clipping and restored the bias offset. Based on the Nikon hackers who restored the bias offset for Nikon Exmor cameras, the read noise at ISO 100, after the offset was restored, was around 6e-. I think the A7s ISO starts below ISO 100, where it's read noise is a little higher, however at ISO 100, I believe the A7s is around 6e- as well...given the similarities between the A7s and an offset-restored Nikon D800, I think that's all Sony did: stop clipping.

No... the A7S has significantly more read noise at base ISO than the D800, A7R, etc. Also, the A7R has an offset and doesn't clip its black. Yet it has significantly more DR than the A7S. The A7S has a different architecture, which explains why its read noise has two plateaus - it doesn't just drop as a function of ISO. It drops, flattens out, then drops again, then flattens. Something about this architecture lends it more measured downstream read noise, which limits its DR. It's still significantly better than Canon, but not as good as the A7R, the D800/810, etc.

One tricky thing is that absolute extreme ETTR can make processing tricky as most standard tone curves end leaving you with poor highlight separation and a lot of stock color profiles are twisted so you can get weird color tints and problems.

Good point. Probably explains some of the odd colors I see with 6 EV pushes

Precisely! You get more room to manoeuvre in. Less restrictions on which combinations to select from to get what you want and need. So it is, indeed, giving more than just technically better rendition on the pixel level, you get that extra artistic freedom to combine settings that's usable.

Now that's just talking too much sense!

Yeah dialing contrast way down and saturation down a little bit is a good idea too, if you want the regular histogram to give a better picture of what the RAW is doing.

Interesting. I've been trying to find a way to get the histogram to better indicate what's happening in Raw, without much luck. Perhaps 'Flat' profile and Adobe RGB on the D810 would be a good place to start. Thanks for the tip.