News Reviews Features Buying Guide Sample Images Videos Cameras LensesPhonesPrinters Forums Galleries Challenges

369

Sony Alpha 7R II: Real-world ISO invariance study

Published Aug 24, 2015 | Rishi Sanyal

The recently announced Sony Alpha 7R II packs a lot of groundbreaking features, and one of its highlights is the new full-frame 42.4MP BSI-CMOS sensor. Sony sensors tend to show very high Raw dynamic range due to low noise characteristics, but we wanted to take a closer look at exactly how little noise the camera itself contributes to images. So we performed our typical 'ISO-invariance' analysis on the camera, but with a real-world scene. Have a look at the results for yourself.

Preface

Recently, we've been testing the 'ISO-invariance' of cameras in a controlled manner, as a means of comparing Raw dynamic range across cameras. Before we move on to the Sony a7R II results, we thought we'd do a quick refresher here.

A camera with a very low noise floor is able to capture a large amount of dynamic range, since it will add very little noise to the detail captured in the shadow regions of the image. This means you can push those shadow regions and make them visible, allowing you to expose your scene for the highlights, and 'rescue' shadows later. Cameras with lower dynamic range provide less of an ability to do so, meaning that pushed files look worse than files exposed 'properly' at the time of capture.

All of this has an interesting implication: a camera with a very low noise floor (high dynamic range) minimizes the need to amplify the sensor's signal in order to keep it above that noise floor (which is what ISO amplification conventionally does). This provides an alternate way of working in situations that would traditionally demand higher ISO settings. In a nutshell, it means that once you've determined your optimal shutter speed and aperture settings, you can forego the higher ISO setting - which really only brightens your file in-camera - and adjust your exposure later.

Why would you do this? To protect highlights. In fact, I used this methodology to achieve the shot taken on the Nikon D810, a highly ISO-invariant camera, below:

We've published this image a few times on DPRreview, because it's a good real-world example of what can be achieved when using a camera with very wide dynamic range.

This shot was exposed for the highlights by keeping ISO near base. I shot in M mode which allowed me to dial in a wide-open aperture and the shutter speed necessary to freeze the action (1/640s). In order to ensure I didn't blow any of the sky, I kept my ISO down at 140. Pushing shadows/midtones 4 EV in post, while holding back the highlights, I was able to tonemap the scene to my desired vision, without tones becoming too noisy. All thanks to the Raw dynamic range of the Nikon D810, and its Sony sensor.

ISO-invariance

In our ISO-invariance test, we do something that may seem counter-intuitive: we use the same aperture and shutter speed at different ISO settings, and then equalize brightness of all resulting Raw files in post-processing. Recall that your ISO setting in your camera is simply an in-camera 'brightening' slider of sorts, so what we're essentially looking at is: what is the difference between brightening in-camera by increasing ISO (and using hardware, mostly analog amplification) vs. brightening after-the-fact in your Raw converter (digital correction)?

Below, you'll see a bunch of shots at different ISOs. Note that all shots X EV below ISO 6400 were adjusted +X EV in ACR. For example, the ISO 100 shot was pushed by 6 EV, ISO 200 by 5EV, and so on. An 'ISO-invariant' camera shows little cost to boosting a low ISO file in post-processing, while an 'ISO-variant' camera (which has lower dynamic range) shows significant cost to performing this exposure adjustment in post-processing as opposed to in-camera via setting your ISO to a higher setting.

As you can see above: there's a very modest visual difference* in noise between shooting at ISO 6400, compared with using the camera's base ISO (100) and digitally pushing later. Why does this matter? Well, for a start, it shows that the sensor is contributing very low levels of downstream read noise, which is impressive in itself. But it also opens up the option to use the shutter speed and aperture value you'd usually use for ISO 6400, while staying at ISO 200 and using a much lower level of amplification and, then, selectively brightening your image later. Using a lower level of amplification means that highlight detail is less likely to get over-amplified and blown-out. And you can see this in the signage: lower ISO settings retain more detail in the bright sign, whereas all detail is blown in the ISO 6400 shot.

Now, we're not saying there's no cost to keeping your ISO low and brightening in post.** We're saying that the cost of a 6 EV push of an ISO 100 shot (vs ISO 6400) is only a mere half a stop or thereabouts in shadows, with almost no visible cost in midtones.*** Meanwhile, you give yourself no less than 6 EV highlight headroom by decreasing your ISO to 100.

Raw Compression

Speaking of that sign, take a look at the unpleasant artifacts around it. With the a7R II, Sony continues the tradition of lossy Raw files, with no option for lossless 14-bit Raw. This isn't going to affect all your photographs, but it can rear its unsightly head at high contrast boundaries of pushed files. The area around the sign in the ISO 100 + 6EV and ISO 200 + 5EV shots have artifacts not present in the higher ISO shots. These posterization artifacts are due to Sony's lossy compression, which you can read more about in Iliah Borg's excellent study here.

We think this Raw compression is particularly a shame, especially when you consider how ISO-invariant this camera is otherwise. As you saw above, the pushed ISO 200 shot has very little additional noise compared to the ISO 6400 shot. This is because of the high dynamic range, and ISO-invariance, of the camera, which means that realistically you can save much of your image brightening for post-processing as opposed to increasing your ISO setting in-camera. However, you'll be limited in your ability to do so because of Raw compression, essentially meaning the camera isn't as ISO-invariant as it could be.

In summary...

As we've come to expect from Sony sensors, the a7R II's high dynamic range means the camera is fairly ISO-invariant. This means you can often save image brightening for after-the-fact, affording yourself (potentially) stops of highlight headroom by keeping analog amplification low in-camera. You can do this by shooting in Manual, holding your shutter speed and aperture at whatever is needed for the higher ISO your low-light situation demands, but then simply dialing down your ISO setting.

Sadly, Raw compression means you'll be limited in your ability to do so, due to the potential for posterization artifacts around high contrast edges.

One thing to note: on many cameras, working in the manner we're suggesting may be awkward, because it can mean trying to operate with a very dark preview (because the camera thinks you're under-exposing when you dial down that ISO setting). But on Sony's newer cameras, you can employ the ultra-flat S-Log2 gamma curve from the Picture Profile menu. This super-flat tone curve ends up representing relatively dark captured tones as mid-tones, which allows you to see the wider dynamic range the camera and our eyes are capable of when shooting in the manner we're suggesting.

Footnotes

* Remember to turn off 'Shading Compensation' so you don't get weird artifacts like this near the edges, even in Raw (scroll back up to the widget after clicking this link).

** In terms of noise levels, there actually is some cost to pushing in post-processing as opposed to performing analog amplification in-camera (leaving aside discussion of Raw compression). In this case, this can be both a good and bad thing. Even with the best cameras with the lowest noise floors (e.g. Nikon D810), we see a bit of a noise cost to performing amplification in post vs. in-camera, and this is due to some residual downstream read noise, likely including some quantization noise due to insufficient ADC bit-depth for the large full-well capacities of such cameras. Part of the noise cost with the a7R II, though, is likely due to an additional benefit of the higher ISO settings on this camera: a higher conversion gain. Bill Claff's data on the a7R II vs the a7R indicate that at ISO 640 and above, a higher analog gain is applied fairly upstream in the signal pathway, which can help overcome the finite downstream read noise that, while incredibly low in Sony sensors, still is non-zero. In this case, then, 'ISO-variance' is actually a good thing, with the camera only truly ISO invariant from ISO 640 and above. Confused? It's a bit much to elaborate upon here, so let's discuss in the comments below!

*** For reference, here's what 1 EV differences in noise performance look like between different ISO settings on a D750, or what a 1 and 1/3 EV difference in noise looks like between formats (D750 vs D7100 in the link provided). The noise performance cost we see in shadows in our a7R II ISO 6400 vs. ISO 100 + 6EV shots appears to be less, to our eyes, than any of the differences you see here, hence our 'merely half a stop or thereabouts' comment.

Tags: alpha, cameralens-features, sony

Gear in this story

Sony Alpha 7R II

302

I own it

967

I want it

I had it

Discuss in the forums

23 customer reviews

Sony ILCE7RM2/B a7R II Full-Frame Mirrorless Interchangeable Lens Camera, Body Only

In stock

Usually ships in 24 hours

$3,198.00Shop now

Cameta Camera

Sony Alpha E-Mount Vario-Tessar T* FE 16-35mm f/4.0 ZA OSS Zoom Lens with 3 Filters + Backpack + Kit for A7 II, A7R, A7S Cameras

$1249.95

Comments

Total comments: 369

I have still not seen why no one mentions the split ADC approach used by Sony on A7RII and other sensors.

I guess it is about video ease/speed vs a full 14/16 bit ADC, but does it somehow effect the ability to do full 14bit loss-less Raw files ??

Anyone know?

0 upvotes

The dual conversion gain happens inside the Active Pixel Sensor (APS) so I doubt this has any effect of the bit depth of any downstream electronics.

0 upvotes

Nothing really special for $3200 body only cam.
The D7000 does just as good, even at ISO-5000.
Just learn how to use your tool.
http://s244.photobucket.com/user/rvrsbnd/media/de8c4ba4.jpg.html

Comment edited 48 seconds after posting

0 upvotes

Great test! Move cursor to the Street-pointer on the word "Was...". Under letter "W" there is a light source. Its going to loose details already on ISO 200. So you can't restore it, if you have no data. So, if one wants to get max from this particular camera, just follow the test's advice - keep it as low as possible, use PP. Yes, this fantastic machine for people with very advanced PP technique, otherwise use something else, 800-810 is great enough. I do!

1 upvote

Look the dual light-pole at the corner, at ISO 100 you can see the air holes on the metal ring around top of the glass, but after 100-400 (depending which one you look) the details are gone. So bright light areas gets severely clipped, but otherwise details are kept very well.

But nothing so special after all.

0 upvotes

Fri13, with some reservations, agree. What I like - 5 ways of OS, wide abilities to use any optics. But DR is still less then Nikon's one. Will see.

0 upvotes

Rishi, you are too aggressive in your replies, and not the first time either. Its obvious what's going on with Sony

2 upvotes

Perhaps Rishi is understandably tired of having to defend what is a completely reasonable review? Or perhaps he's tired of newbies and laypeople telling them how to run DPR and what they "should" be doing.
I know I am.

2 upvotes

This is definitely a great article. Keep them coming!

1 upvote

How to do a +6EV push in ACR?

0 upvotes

I know of two methods.

1st and easy: make an off-image gradient, boosting exposure by +4EV. Then the exposure slider effectively works as -1EV to +9EV.

2nd and more complicated: Export image to PS (as 16 Bit Tiff) and convert to 32 Bit. Save and back to LR. Now the exposure slider magically supports a -10EV to +10EV range.

1 upvote

@falconeyes: No way! Giving this a try right now (your option #2). If that works, you'll have saved us a lot of headache :)

Does LR preserve its aggressive highlight rolloff with this method? One issue with taking images into Photoshop is that exposure adjustments are applied without roll-offs, so highlights clip easily.

Personally, I think ACR needs to turn up the volume on most of its sliders to cope with the incredible dynamic range of certain recent offerings.

2 upvotes

We can run some experiments. My #2 assumes that the RAW is preprocessed to have no clipped channels. Also, I assume that the 32 Bit file is reimported into LR. No exposure is altered within PS!! The LR rolloff behaviour should remain the same.

But I confess it is a while ago I tried this the last time ...

> ACR needs to turn up the volume on most of its sliders

IIRC, LR increased the exposure variance from 4 to 5 EV.

1 upvote

@falconeyes I had no idea that the slider will magically have 10EV+/- when converted to 32-bit! But I did notice a strange change. When I "Edit in photoshop cc" and I convert it to 32bit mode then then save it which naturally adds 32-bit TIFF in LR and for some reason, the image looks more like the default JPG (more contrast/saturation) and is overall brighter than the default on NEF. In fact, the TIFF does not match the TIFF in PS CC at all. But I was able to use the EV slider and get far more details in shadows than I could in default NEF. Definitely something is going on. I am just stunned how much potential there is. I am using Nikon D5500

@Rishi it would be nice if you could report your findings with falconeye's methods.

0 upvotes

A 32 Bit Tiff file has arbitrary default brightness (set when being converted). The initial brightness when imported back into LR is to be ignored.

BTW, beside 32 Bit Tif, 32 Bit linear DNG does the same trick. It can be created by DNG converter (IIRC) or by LR 6's new HDR merge feature. Maybe, HDR merge would work with two identical RAWs too ...

1 upvote

That's a very useful tip. I need to play with this sometime.

0 upvotes

I now played with the possibilities a bit and this is the result ...

First, let's recall the methods:
#1 Off-Image Gradient+4EV + Exposure
#2 Export to PS, save as 32 Bit
#3 Duplicate RAW, merge both into one HDR (LR-Shortcut ^H)

In #2 and #3, exposure sliders support a -10 to +10EV range.

Findings:
Methods #1 and #3 yield identical results. #1 is easy and remains my preferred method. Rishi, what do you use at DPR?

#2 works and yields similiar results. However, unlike the 16 Bit tiff file, the 32 bit tiff file has much more default contrast. This makes it a bad choice for one to one comparisons.

The reason why #2 doesn't work as expected is a bug in LR: PS saves 32 Bit files with gamma=1 (whatever be the profile) but LR treats it as if having gamma. To work around, one would have to use a linear color space supported by LR and convert to that in PS before converting to 32 Bit.

Happy re-exposing :)

0 upvotes

Thanks for the additional info. :)

0 upvotes

Here is an idea, why doesn't DPR just do a review of the camera and be done with it?

Instead we have extensive discussions of the technology of this particular camera and "reviews" of other cameras that fall far short content-wise of what Phil Askey used to put out years ago.

0 upvotes

So when we forego really, really technically detailed analyses for a product for a quick review (E-M10 II), b/c it's more an iterative update than a whole bunch of new technology that needs to be tested/addressed, you complain the reviews fall short.

OTOH, when we go really in-depth, & publish one piece at a time b/c each piece is so in-depth that it requires its own treatment, and so that we can update you on findings while the camera is still relevant, and then you also complain.

If you could carefully spell out exactly what it is you'd like us to do, we'll try and address your feedback, b/c as it is, I'm honestly confused.

6 upvotes

Rishi
OMG...Weekend. Finally got to dig into the S Log 2 that you explained some in the RX100m4 review. Also a new feature in the A7rii.
This picture profile thing is making a whole new level of adjustments available to photographers... Real world potential is HUGE... We need a whole separate article or series of articles just about the adjustment features in picture profile. I am testing today. But it will take weeks just to understand the many areas of potential and learn how to apply. Don't listen to Dave. We need more technical articles that help us learn to use the cool new features coming our way.

1 upvote

So Rishi please keep up the good work.

Comment edited 1 minute after posting

2 upvotes

Tech/detailed reviews simply make some ppl's HEAD HURT !! Trying to do something the brain is not capable of maybe ?

Keep on with as much tech detail you can, but add a very simplistic 'Executive Brief' for those whose heads hurt !! ;)

Ouch, me head hurts !! ;)

1 upvote

Yes, I like the idea of an Executive Summary (i.e., "tl;dr" version) - but perhaps that's what the already included Conclusions is about. Maybe feature that Conclusions section up front for those that don't want to skip directly to the end using the pulldown menu?

0 upvotes

There are dozens of 'Quick' / 'Executive Summary' / 'Feeling' type reviews. I ignore them all. I want highly technical analysis when looking at buying a camera. DPreview is one of the few sites that provide that.

0 upvotes

Regarding phase detect sensors embedded on the image sensor: I assume these phase detect "pixels" take real estate away from image pixels? Does Sony interpolate image data for these "holes" in the image array?

3 upvotes

Yes, it gets interpolated just like on the Olympus E-M1 does it.

But it is just a single R, G or B pixel, that gets then calculated from surroundings without data loss.

Only pixel peepers are worried about it, even when there are 399 sub-pixels lost because of that, from over 42 million of them.

http://www.imaging-resource.com/PRODS/olympus-e-m1/ZTECH_PDAF_PIXELS.gif

Comment edited 43 seconds after posting

0 upvotes

The
DSC09021.arw
is the ISO 100 Image. I just ran it through originalraw. All i can say about the original JPG result is, that the image is massively underexposed. Maybe its underexposed by 7 stops. Boosting that image by 6 to 7 Stops and complaining afterwards doesnt make any sense.

Who on earth takes pictures at EV-7 GENERALLY ??
I never do.
As a photographer its your job to expose the Sensor of you digicam CORRECTLY to get the maximum SNR.

I think Sony did very well here, providing
40 megabytes for each ARW file (lossy)
instead of
70 megabytes for each ARW file (lossless).

Lossless RAW with 70 megabytes may be only useful for a few boosting brightness by EV+7.

But the majority of the photographers can save a lot of Space, and most of them, never take pictures at EV-7.

Lossless ARW with 70 megabytes filesize may be a welcoming option, but it should be optional, but i would never use it.

Comment edited 2 times, last edit 1 minute after posting

1 upvote

I would suggest Sony giving ppl a choice of lossy or lossless raws.

As to why Sony can not do a lossless but compressed raw still baffles me, as don't other makes offer this ??

4 upvotes

@ www_zeeshan_de,

unfortunately, you don't have a valid point.
First, you seem to miss what isoless is all about. Read up about it, it will benefit you in your photography. Rishi does the right tests here.

Second, Nikon D800 14 Bit lossless NEFs are the same size as Sony lossy ones. Sony destroys image information for no reason. Lossless compression is as old as ZIP and older...

And BTW, there are much better ways to do a lossy compression. No measurable loss in image quality and much smaller output files. No vendor does it ATM, but the methods exist. Maybe, I should write about that ...

Comment edited 2 minutes after posting

0 upvotes

@falconeyes re:lossy compression alternative algorithms - no measurable loss or no "perceivable" loss? Got links?

Sony probably thought their scheme was imperceptible too until DPR formulated a pathological - though not unreasonable - example.

0 upvotes

@steve_hoge the algorithm is unpublished own work. Like some lossy compressors, it makes use of photon shot noise statistics. However unlike any existing approach, I rescale the statistiscs to hold true at all spatial frequencies. Something the other lossy compressors fail at miserably. Also, I add any noise I removed at compression when decompressing. Therefore, two separate reads of the raw file yield two different uncompressed images. However, against a third, stored w/o compression, no method could know which one was stored which way.

0 upvotes

@www_zeeshan_de, of course, you don't need to shoot at EV-7 (I don't think there is any camera with that much range in EV). The point of the article is showing how it makes little difference between shooting with a given high ISO in camera and pushing the ISO in RAW editor from an image taken at ISO 100 if you're using an ISO invariant sensor. When you shoot at low ISO despite being underexposed by a few stops and develop in an editor, you could preserve highlights better.

With that in mind, using the ISO invariant sensor, it's a good idea to expose for the highlights and push the shadows in RAW editor and not worry about the chroma noise that would have popped up if for an ISO variant sensor.

The technology behind the ISO invariant sensor is exactly why I switched from Canon to Nikon. I do see huge differences in shadows when pushed between Canon and Nikon. And DPReview confirms that in several tests.

Comment edited 3 minutes after posting

0 upvotes

Wow is it just me, or are these posts getting so technical that you need an MIT engineering degree just to keep up? where are these posters coming from. I though i knew what was going on with photography, i'm at a loss.
I think that a review should be done without revealing the camera and see what happens, then reveal the camera at a later date, kind of like a non biased camera review, and posters should give the camera or whatever item a rating of 1-10.

Comment edited 14 minutes after posting

4 upvotes

Just You !! ;)

2 upvotes

Agreed...time to go take some pictures...that process hasn't changed has it?...

0 upvotes

Do you shoot JPEG only? Then this is not for you, although you can use extended dynamic range option in camera to do the same thing.

New cameras allow ypu to underexpose to retain highlight detail. The very dark shadows in the raw file can then be brightened without much loss of quality. This can be of use in many real-life photography situations.

0 upvotes

@Mike, apparently it has changed. We are now supposed to let the camera identify the eyes for us, all you do is blink twice and the shot will be had. By 2017 they will be shipping a free French photographer with the camera, that way you don't have to do anything, you just tell Pierre what to shoot.

Comment edited 36 seconds after posting

1 upvote

DPReview, please forward this article to Sony. Right now.

I've said in a previous article that this was a "perfect example on the RAW compression subject". Glad to see you've picked it up again, with a more in-depth explanation.

1 upvote

@DPR don't you think these magniloquent articles for Sony are a little over sonorous?

I like what Sony is doing, but I also appreciate what other manufacturers are doing. Keep it balanced please!

4 upvotes

Why don't you read it before commenting?

6 upvotes

@Rouse... it beats magniloquent comments wouldn't you say?

I get your tongue-in-cheek comment, but the fact is the article goes into depth on the single biggest flaw that the A7Rm2 has. This is important as it shows Sony that potential buyers care about this flaw in what could very well be the best consumer camera ever made. I am an enthusiast but the only reason I haven't bought this camera is because of this issue. Maybe it's fixable in firmware, maybe not.

You might not care, but I guarantee there are plenty of us that do. Furthermore, DPR can ONLY keep it balanced if the "other manufacturers" are keeping up with new and exciting product development - which clearly, they are not.

3 upvotes

To what degree is the non lossless raw to do with Sony's split ADC approach ? Is this some strange complicating factor?

As to other manufacturers keeping up, why are those Toshiba/Aptina sensors used in Nikon cams pretty nice as well OR that FujiFilm are offering a 'real' electronic/global shutter ?

4 sec. to 1/32000 sec, faster than any Sony cam shutter, NO ?

Odd Sony does not have it all, do they ?!!

Objective thinking cleanses the soul !?

0 upvotes

I really wanted to edit the comment because it needed clarifying, but couldn't figure out how to on this mobile devise, then work took priority. Now clarification would just be seen as little more than a pathetic backpedal. So lets just keep talking about the only camera maker on the market doing anything because apparently the fools are believing it!

1 upvote

So, it's like decades before... if you buy a Sonyproduct you get most likely one of the best techniques, but with a horsefeet on it. I bought my videocam and got Memorystick only back in the 90s, I got my Bravia and had to get Sony to Realworld connectors just to plug in my peripherals. This time we pack together a good sensors with a lossy RAW ;)

6 upvotes

It is clear that ISO 100 is not sufficient enough light for low light shooting. ISO 800 or more is much better result I tried when I download your raw file. The ISO 100 the shadow is clipped so badly it is impossible to recover nicely and end up with rear end ugly artifacts. So I tried the ISO 800 and had no problem recover the shadow with ease and no artifacts were observed. I did not need exposure push at all, it looks good. It demonstrates that ISO 100 is useless anyway for low light, I prefer higher ISO as a result for better recoverable of shadow in particular and less noise in shadow with lower recovery amount so it looks much cleaner. That's for low light photography. Interesting post though. Thanks for showing that. I think it is totally unnecessary to have 14 bit raw file usually if I were to use higher ISO for low light especially starting from 800, that's more than satisfied enough.

4 upvotes

Rishi, thanks for the fine article. One question: Do you expect the Studio Comparison Tool for DR will be changed from fixed ISO with underexposure by faster shutter speed (plus corresponding PP pushing) to your method here of fixed shutter speed with underexposure by decreasing ISO (again, with corresponding PP pushing)? Thanks.

Comment edited 24 seconds after posting

0 upvotes

Rishi,

I think you've got about 75% of readers misunderstanding what you are trying to convey here. Maybe another intro paragraph to the article making it more simple to understand? Your mix of lossy raw and underexposure discussion is confusing.

3 upvotes

Was lens shading correction turned on? There seems to be some strange artifact around the corners in the lower ISO files that contributes to the noise. When I have shading correction turned on the raw files from my NEX-5N have a similar thing going on. If it's not that perhaps it's the lossy compression.

The performance seems very good for a 42mp sensor and 5 stops should be way more than most people would need to push shadows anyway, I often go about 1 stop and the shadow slider to full with files from my NEX and that's enough to lift almost black parts into midtones, when the same adjustments are applied to the iso 100 Raw it's very clean indeed, as good as or better than my 16mp NEX which is a pretty big achievement for a 42mp sensor!

2 upvotes

As a service to the community and a research project at the University of Kentucky, I've been working on a free program to credibly repair the artifacting in ARW2 files. I asked for aligned training images differing only in exposure at http://www.dpreview.com/forums/post/56345899 , and have gotten some from various cameras. However, I also can use the above type of ISO sequence for training. I'd like DPReview's permission to use the above images for this purpose and, with appropriate citation, in any scholarly publications on the topic.

PS: The repair algorithm isn't quite good enough to be posting yet... hopefully within the next week or so I'll post a link here and/or in the FF Sony E forum.

8 upvotes

The tool still isn't ready for posting, but here are two crops from the above ISO 100 raw data that show the repair quality is fairly good:
http://www.dpreview.com/forums/post/56389545

1 upvote

Thanks for pointing us to your work. Interesting.
According to Sony, may soon become obsolete (when they fix their ARW format), but still an impressive piece of work.

0 upvotes

This is good information and not generally seen. Excellent work, Rishi. DPR, give this man a raise!

8 upvotes

Such tests have been posted by forum members since the Sony A700 and A900 in the Sony A Mount forum, and Nikon forums readers should also be familier with similar work. Nothing new. Good to see that the Dpreview staff have started testing real life dynamic range though! :-)

Comment edited 7 minutes after posting

1 upvote

@Magnar, you are right, but before Rishi, the technological expertship of DPR staff sometimes lagged a bit behind their readers ;) No more :)

Comment edited 14 seconds after posting

1 upvote

@Falcon, you wouldn't know it by reading the forums. Sift through E mount sometime, if you believe them, there is NOTHING wrong with Sony files, it's a myth. Compression issues are not seen in real world shooting. Right?

0 upvotes

Actually, the vast majority in the E mount forum accept that there is an issue. The discussion is not wheter the artifacts exist or not, but how this affects real world photography. For me, something like a slamming mirror is much more harmful that lossy compressed raw files when using long telephoto lenses and telescopes, and also when I am doing macro stuff. Depends a lot on how you are using Your camera. And yes! You are right! The compression artifacts are not seen in real life shooting unless the files are pushed a lot! For those of us who are using low ISO/pushing techniques for night scenes/astrophoto know very well the risk. We rarely see the artifacts. Still they exist! ;-)

0 upvotes

I love it when people say "SLAMMING MIRROR!!". I get this mental image of a dump truck hitting a wall at 100mph. Back in reality today's mirrors are often dampened very well. My K3 for example is very quiet and you can hardly feel any movement. My GX7 had more internal vibration with just a shutter.

As for the artifacts they are often and important enough for Rishi to write up an extensive article on it no?

0 upvotes

For long telephoto/telescope/microscope and much macro work mirror shake actually causes VISIBLE loss in sharpness. I often do such work. Even shutter shake is a problem! There is a significant and visible difference even when locking up the mirror and using ordinary shutter vs Electronic first curtain shutter. When not locking up the mirror and using shutter speeds in the range of 1/2 to 1/250 sec, sharpness is seriously affected for the work mentioned. Just test this yourself - easy, and the result will speak for itself!

Comment edited 38 seconds after posting

0 upvotes

Check out the two tree trunks on the left hand side. Notice the one without a light behind is fine, the one with the bright shop window behind is barcoded.
Who would come up with an algorithm that compresses the more noticable details in an image? Its a ludicrous compression scheme for a photograph especially when the size reduction of the file is so small.
You do need a high dynamic range scene with high contrast edges to see it but the camera would be capable of tackling these scenes without the compression issue. Took the shine off for me when I found a very strong example of the issue in the first week with the camera (visible with far less of a push than here).

6 upvotes

Take it down to print view and it is still noticeable to.

2 upvotes

"quantization noise due to insufficient ADC bit-depth" for D810, would you suggest Nikon to use 15/16-bit ADC to improve quality for their new cameras?

What is the minimum ISO of D810 that is ISO invariant?

How to do a +6EV push in ACR? I could only do +5EV.

Thanks

1 upvote

Would you mind sharing your optimal ACR settings? I downloaded the ISO 100 file and no matter what I did, the image was riddled with noise by +2 when looking at it 3840x2160. I can't post an image here, but did you realize there was a bum laid out in the store's entryway?

2 upvotes

"Riddled with noise", as in more than what's shown in their images? I can't get +6 exposure in ACR, but with +5, -50 highlights and Tungsten WB, it gets me almost exactly to what's shown in the images above, including noise. All other settings were default, including NR (25 Color, 50 Detail, 50 Smoothness)

1 upvote

I'm with Rick, same story on 4K screen so easily can see how noisy it is. Not sure whats up with this?

1 upvote

Well there's a probably homeless person in the doorway. No need to denigrate when their story isn't known.

1 upvote

bum /bəm/ noun North American

1. a vagrant.

He's looking pretty vagrant to me.

1 upvote

I picked up the phrase 'riddled with noise' from a Nikon/Sony troll describing 5DsR images on the Canon forum. This image is very similar to the Canon image so...

4 upvotes

judging by peoples lauding RAW strengths (Sony ARW) for pp
if the ARW shadow recovery benefits differ only by a little bit from Sony ooc jpegs

here:
http://www.dpreview.com/forums/post/56372417
http://www.dpreview.com/forums/post/56372472

wouldn't Sony ARW (cooked) RAW just be a slightly (mathematically) stretched (less compressed) kind of lossy JPEG (can manipulate more than 8-bit jpegs) whilst not officially a jpeg (it isn't 8-bit; therefore not jpeg by definition)? right?

are there lessons for RAW-only shooters here?

I can't be sure why ooc jpeg would be so high contrast (why darks too dark in full daylight) even with the sun (well above horizon) properly exposed (far brighter than a night time store lit display, right?!). why not just dial contrast down in camera so ooc jpegs aren't rendered unrealistically dark?!

Comment edited 6 times, last edit 14 minutes after posting

0 upvotes

No the two aren't comparable (RAW & JPEG). I shoot weddings with an A7 and I shoot indoors so at high ISO like 1600 and more. Whenever I clip the highlights (which happen very very often) with RAW I can pull back a pretty substantial amount of detail.

JPEG has very little headroom to pull back clipped highlights. A one stop clip is enough to destroy any detail in the highlights when brought back. With RAW I can pull back detail from 2 to 3 stops over exposure.

4 upvotes

I think you miss my point.
true unaltered RAW show exact highlight clipping (and sunk shadow clipping)
cooked RAW overlays a 'rendered contrast' or 'rendered gamma' that should not be there to begin with.

darks that aren't really at black point are overexaggerated with too dark darks (not really black blacks)

brights that aren't really at white point are overexaggerated with too white brights (not really white whites)

initially digital RAW images were very flat, with unaltered gammas, regardless of over or under exposures. this confused users used to film era contrast found in positives (slides) and negatives (print papers, etc)

today, dcam mfrs (and software app mfrs) have opted to precook the look of RAW with altered (exaggerated) contrast to mimick film contrast

keeping in mind, FILM contrast NEVER matched reality, ever.
I.e. shadows never look pure black in broad daylight, indoors, or night to our eyes, it was a phenomena of film, which digital is mimicking, to acclimate users

Comment edited 2 minutes after posting

0 upvotes

basically, flat RAW (true RAW) is returning, initially for pro Cine users, whilst flat RAW is SLOWLY returning for prosumer still users in terms of flatter contrast/gamma profiles.

adjustable previewable live contrast/gamma (DR rendering) for in camera jpeg engines have always been around, even though not all camera mfrs offers it fully (or partially, or none at all).

Comment edited 28 seconds after posting

0 upvotes

I'm glad to hear there is something to moan. Even if it is only visible if I look on less than 1% of the picture on a 4K monitor. :)

If your doing ISO comparisons then use the 640 instead of 400 and 800. There are a lot of researches in dpreview forum that shows, where this camera has best dynamic range. And for such tests it's just stupid to let 640 out.
If you are pushing EV then try this in LR and C1. I had very good results in C1 with pushing such files, sharpen/noisereduction and using moiré at the end to get rid of the posterisation.

Thanks for the tests and suggestions Rishi!

1 upvote

There's a lot of confusion in the comments. People should read DPReview's "Sources of Noise" series from a few months back to get on the same page.

1 upvote

And just so folks have it here:

Hope this helps.

0 upvotes

Interesting test. ISO invariance sensor is almost there. But for best result, i would use iso800 for low light scene. It achieve equivalent clean shadow area as ISO6400, and definitely better than the lower iso comparison.

2 upvotes

Did anyone else notice the white line over the womans chest like some bad photoshop skills showing through?

3 upvotes

Are you talking about the horse riding photo? I don't see it

0 upvotes

Please look here:
https://www.dropbox.com/s/4kaavs1287emzzg/dpr_hairline.jpg?dl=0

This is not a monitor nor a GPU issue with me :)

And after a closer look i can clearly see that the editor sharpened everything but the area of the head. The fabric is less sharpened from the line upwards towards the head. I would say DPR really needs to hire a better skilled PP guy or girl. :p

Comment edited 3 times, last edit 7 minutes after posting

4 upvotes

Thanks, it is there and strange.

0 upvotes

Yes, I see it. Also, look at the dark area on her upper left arm. Looks more like gangrene than shadowing. And, do we have to point out that the image is riddled with noise?

0 upvotes

Whoa, weird. I didn't do any masking with this image, so I'm not sure where that's coming from. If you look at where it connects to the top of the top, though, it appears to continue into the fabric near her shoulder. Wonder if it was a loose thread that was oversharpened - I did use masked sharpening.

I'll go back and have a look at the original.

Rick: the image isn't riddled with any more noise than you'd expect from an ISO 2200 full-frame shot, which was the ISO required for a 'proper exposure' given I was limited to F4 (wide open) and 1/640s to freeze the action. So I don't see your point.

6 upvotes

It's clearly on, or part of, the dress.

0 upvotes

it's part of the dress stitching, blew it up 400% it also goes up the top of the dress shoulder

0 upvotes

No it's not. Noise levels and sharpness differ between the fabric above the line and below.

1 upvote

"Noise levels and sharpness differ between the fabric above the line and below."
Yes --especially at the left shoulder/lace -looks almost like another layer of gauzy material, there. Very interesting.

Comment edited 5 minutes after posting

0 upvotes

Interesting article about a subject that deserves more scrutiny. Thank you! But I am afraid that your analysis is superficial and your conclusion faulty in this instance. There is more than one source of noise, and one of the purposes of analog gain (increasing ISO) is to reduce quantization noise from the A/D converter.

The place where you put the magnification box shows blooming and compression artifacts in the back-lit sign (although the lettering can be recovered up to ISO 3200 in Raw PP); but it is not a good demonstration of shadow noise, as the "dark" area is lit by the streetlight. Move the box to a shadow area (for example, the building top left that is almost in darkness), and you see very significant noise, decreasing steadily as ISO is increased up to 800, with minimal improvement beyond.

6 upvotes

One step ahead of ya... I'd already mentioned exactly that in footnote #2.

That is: all about quantization noise being decreased with increasing ISO b/c of the lower effective FWC and therefore the lower photoelectron/DN conversion factor, which means lower quantization noise at higher ISOs.

And, fair enough, really, really, really deep shadows may have more of a cost, but if you're still using those shadows after a 6 EV push, then... so in terms of real-world relevance, I'd say there's only at most a 1 EV noise penalty in shadows, and probably actually less. Much less of a noise penalty (approaching zero) in midtones, and mind you these are midtones after the 6 EV push.

Meanwhile, the gain in highlight headroom is massive.

4 upvotes

Not just the very deep shadows -- all the shadows. And you are not seeing "highlight headroom" in the unreadable sign; this sensor seems to have a serious blooming problem.

1 upvote

Chris, do you mean this sign?
https://s3.amazonaws.com/f.cl.ly/items/3g0I3Q2K223c3B1l0Y2l/Screen%20shot%202015-08-25%20at%202.10.51%20PM.png

0 upvotes

Good point Iliah. The exposure is the same for all the images so we're not seeing blooming; and the Raw histogram of the sign @ ISO 6400 looks very good. Still, what Rishi refers to as "massive highlight headroom" looks to me simply like 6 EV of Raw underexposure at base ISO, and lots of extra shadow noise up to ISO 800 (i.e. I don't see ISO-invariance).

1 upvote

Sorry -- Raw histogram of the sign starts saturating at ISO 200. But looks well-behaved at ISO 100.

Comment edited 2 minutes after posting

0 upvotes

Chris, IMHO much depends also on how clean a converter can push, and is it good enough for 9-11 stops of tone mapping whilst preserving full details in highlights. Magic Lantern files with dual ISO are very interesting if one is not going to blow them up unreasonably.

0 upvotes

I think the lossy compression is something Sony decided to do to maintain high frame rate given the high pixel count.

3 upvotes

Would then seem to be easy and logical to have a lossless mode for low frame rates. Storage keeps getting cheaper.

0 upvotes

It's possible it's not that simple. May be other consequences. Everything always turns out to be a little more complicated than it seems like it will be going in.
Think Rishi is right (again) on the proposition that you can shoot darker to save high lights and bring up the details in post. BTW thanks Rishi. Been doing that past few days and it has worked.
But I'll bet this is different, camera to camera and picture to picture. because the sensors are different, and also how the nr is applied and also how the magnification is accomplished. .

0 upvotes

I think that the "compression" happens already in the sensor by design using fast/low precision differential ADC's. This allows to feed the image processor and the many hardware pipelines with compact and handy 128 bit blocks of data. I guess that this saves a significant amount of silicon, (battery)power and bandwidth (hundreds of MBytes/sec) through the various hardware and software pipelines. It allows for high refresh rates too. I think that changing that requires a complete hard and software redesign needing faster components and more power/bandwidth.

11 upvotes

Foroa: thanks for the insight, that's very interesting.

Average User: exactly right, it varies from camera to camera. Generally cameras that have very low downstream read noise are fairly ISO-invariant, as they have their ADCs on the imaging chip and digitize the signal very early. There can still be some benefit to analog ISO amplification though, especially when you have large pixel wells (read: full-frame sensors) whose photoelectron capacity is more than the bit-depth of the ADC. When this is the case, not every photoelectron can be counted at low ISO settings, which leads to quantization noise.

One way around this is to to go to a higher bit-depth ADC. The other is to actually use smaller pixels, since then the full-well capacity per-pixel is lower, which is then potentially then better matched to the maximum value of a 14-bit ADC. This is why, I think, the a7R has higher base ISO dynamic range than the a7S, b/c it has smaller pixels.

3 upvotes

I think that the low power design (combined with noise cancelling at various levels) is precisely in the heart of Sony's technology (and competitive advantage) that made it possible to make such compact cameras. Why you see never a technical discussion on that at Sony.

4 upvotes

But Nikon uses the same sensors? They do not have this issue? I don not use either brand but I am interested in this sensor technology.

0 upvotes

There are two parts to the Sony compression. The first - tone mapping - is probably not detrimental to image quality at all, given that Poisson noise (shot noise) provides sufficient dither where tone compression exists.

The second part - the delta modulation - is a problem that shows up when deep shadows are lifted.

Hopefully only the first type of compression is on the Sony ASIC, opening the door to a modified lossy compressed RAW format that does away with the delta mod component which is what causes problems.

0 upvotes

@mdcromer: I've always wondered about this. Does shot noise provide enough dither? Say you have a signal of 16. At ISO 100, say FWC is ~64k e-, which means you can only count every 4 electrons with a 14-bit ADC. That signal of 16 has shot noise ±4, which means you really only need to differentiate, say, 16 vs 12. 16 would be recorded as 4, while 12 would be recorded as 3 (since you're only counting every 4th photoelectron at base ISO). But then if you add an additional level of imprecision on top of that by only recording every 2nd digital step when applying the cRAW compression curve (slope = 2 for input signals up to 2000, which gets recorded as 1000), wouldn't that mean that 4 is recorded as 2, and 3 as 2 as well, meaning ultimately you can't actually distinguish 16 vs 12 photoelectrons in this scheme?

Or am I just approaching this wrong?

I believe Jim Kasson suggests that the ADC itself is driven at 13-bit precision.

Thanks for any insight.

0 upvotes

At the low electron count domain (which is where you would be if your signal is 4 counts), there is no tone map compression applied.

Tone map compression is only applied at higher electron counts, ranging from none in the deep shadows, to maximum at the brightest highlights.

Also, my understanding is that, at least at in the deep shadows, modern sensors are much more efficient than one count per four electrons - more like 50%+ efficiency. But perhaps I am confused there.

0 upvotes

On Nikons, there may be a better way than Manual mode to take advantage of ISO invariance. You can go to Aperture mode and set Auto ISO to maintain a chosen shutter speed as a minimum -- as long as your max allowed ISO is not exceeded. This has the benefit of guaranteeing that you get your shot without going too many stops away from correct exposure.

At the same time, you can set Highlight-weighted exposure so that significant highlights are never blown. Then it's just a matter of bringing up the shadows.

This gives you a little more versatility and automation without much loss of control -- good for situations where you have to shoot fast.

0 upvotes

Yup, this is why in settings like receptions, I set my Nikon D810 to Aperture Priority, set the Auto ISO minimum shutter speed to whatever is needed to freeze the action, then dial in something like -3 EV exposure compensation, since I know that, given the dim lighting, my ISO is going to be up in the 6400 or so range anyway. And that point, I know there's little to no cost to reducing that ISO down to, say 800, giving myself 3 EV highlight headroom.

Which is certainly nice :)

At the same time, I know I'm probably never paying a shot noise cost, since the -3 EV exposure compensation is really only ever adjusting ISO, not aperture (which I set) or shutter speed (which is probably hitting the lowest limit most of the time anyway, since I'm in dim light).

I also shoot this way in M mode with Auto ISO and exposure compensation a lot as well. It's wonderful b/c I rarely get blown highlights on faces, or the ugly posterization you get when colored uplights blow.

3 upvotes

I like that tip, thanks. BTW this is one of the advantages of an Optical Viewfinder as you can still see what you're shooting (although you can often turn off exposure simulation in EVF cameras, I'm not sure how well that works across models).

With the Canons if shutter speed is important I just tend to go fully manual at an ISO-invariant ISO (typically 800) and just sort everything out later. It can be a pain reviewing what you've got of course, especially if focus is critical. It's annoying staying quite a bit longer to be sure you get the shots and getting buried in them. I guess the answer is to use the raw processing tools in the cameras, but that does take a while.

Comment edited 2 minutes after posting

1 upvote

"...a camera with a very low noise floor (high dynamic range) minimizes the need to amplify the sensor's signal in order to keep it above that noise floor (which is what ISO amplification conventionally does)"

This is as wrong as it could possibly be. Amplifying the signal does not have the effect of elevating the signal above the noise floor. The signal is amplified to bring it into the range that the A/D convertor expects. If the signal is too weak, too much of the signal will be below the threshold at which a non-zero quantization value will be recorded. A low noise level simply means that at whatever amount of signal gain is applied, the S/N will be better. That's all. Your entire approach to the subject is based on an understanding that is patently incorrect. Your article is just plain bizarre.

2 upvotes

Depends on what is the source of noise.

2 upvotes

Er, no. Why is there a programmable gain amplifier (what varies with the ISO setting) then? If it were just the range the A/D converter expects, surely there'd only be one amplification setting, and therefore only one ISO, right?

Read any textbook on image sensors (for example this one to understand the concept of stages of amplification and how it relates to dealing with noise and maximizing SNR. Yes, part of the amplification is to fit the voltage range into what the ADC expects (of course, you also select a proper ADC based on the ranges generated by your electronics), but a huge purpose of the amplification is also to increase SNR. And a large part of this is decreasing the effects of noise on your signal via amplification.

Also might I suggest you reconsider the manner in which you provide feedback & engage in 'discussion'?

7 upvotes

Interesting article. I think the possibility of a firmware fix for the lossy compression is highly optimistic, if it was that easy they would have just done it, and have no incentive to have this particular issue with an otherwise highly impressive camera.

Meanwhile the resistance to a7RII articles which has developed recently is also quite interesting, I think its more of a resistance to people thinking they are constantly being 'sold' this camera due to the large volume of highly enthusiastic articles and forum posts and references, even though of course DPR is just reporting on the most interesting industry offering of the moment. I like the read, although would also be interested in seeing more in depth articles about the Canon 5DSR, what with it being the highest rez flagship camera of the biggest DSLR camera seller = highly relevant to current landscape photographers (and this from a Nikon user).

6 upvotes

Just read the Ming Thein semi-review of the camera, he's a sensible guy I was talking to a while back, and knows the business well. Seems to be several groups of tech loving amateurs and pros raving about the advancements and loving the experience but also several saying that its less than the sum of its parts, that there are several great aspects (especially video) but that as a whole setup for extended shooting it doesn't do it for them.

0 upvotes

Can somebody clue me in on the use of Picture Profiles (like the S-log mentioned in the article) for still images? I assume using these doesn't affect the raw file. Can these profiles safely be used to get a better preview in contrasty situations, like bright sun and shade? I've been using A7 cameras for a while, and never knew about these. I don't see threads about them when I search. Any sources, or just advice?

Thanks,

--d

0 upvotes

Sure, you can read a little about it in the section titled 'S-Log2' in our RX100 IV review here. The end of that section then links off to another more in-depth treatise on the concept within the same review.

Hope this helps. It's really quite clever, and makes shooting in the manner I propose in this article here more practical.

With the a7R II, when the camera says 'ISO 800', its hardware-level amplification is really more like ISO 100, so you know the camera is not throwing away much (any?) dynamic range, as cameras have the most dynamic range at base ISO. The ISO 800 setting the camera automatically jumps in to just ensures the camera purposefully underexposes the image by at least 3 EV, giving you at least 3 EV of highlight headroom.

You can just use this mode to ETTR though (dial in as much exposure as you can before you see highlights clip), & then dark shadows are rendered visible by the extreme tone curve.

1 upvote

Great info. Thanks.

0 upvotes

Another great article, looks like +4 EV is the max post processing boost to avoid the compression artifacts. And results in the best compromise between dynamic range and noise.

0 upvotes

ISO 100 and 200 are clearly worse than ISO 400+ I would stick to making no more than +3EV boost in general. It is as it has always been, Nikon gets more out of a Sony sensor than Sony and clearly have better engine. The D810 trounces the A7R at high ISO and is more ISOless than the A7R. You can't really tell at all with the Nikon if it's been pushed +5-6EV. Hopefully Nikon will get this sensor in the D900.

6 upvotes

According to bclaff and Jim kasson's finding, A7Rii has similar low ISO dynamic range as D810, its high ISO DR is beyond the reach of D810.

If Nikon get hold of this sensor, lossless raw and native ISO64(if possible) could squeeze a little more on low ISO, won't be much.

I think low ISO DR is near the 14bit limit

2 upvotes

I think so too. But don't Jim and Bill's findings indicate more like ~13 EV pixel-level DR? Is that b/c the a7R II is not really quantizing 14-bits?

I haven't verified with my own SNR analyses yet, but my hunch from some shootouts so far is that the D810 still has more base ISO dynamic range. Its pixel-level DR is pushing 14EV (~13.8 or so) in my analyses.

0 upvotes

To me it looked as if ISO 800 was the sweet spot. So why bother with ISO 100 through ISO 400? Personally I think this particular test and the emphasis DPR places on it amounts to bunk. Know your camera, bracket your shots, and stop worrying about "shot noise, read noise and ISO invariance" because in real life all that matters is getting the shot.

5 upvotes

I would like to see how Sony will fix the lossy compression issue with the already slow processor and limited battery capacity. I am not going to hold my breath for that one.
That together with the lack of cheap compact fast primes makes it worthwile to wait to see what the others are up to next.

5 upvotes

The processor will actually be taxed less if it does not need to run compression routine prior to saving. Battery life could actually improve. The great thing about this so called issue is that Sony can easily issue a firmware fix to make this camera near perfect. Buying into anything else (m4/3, APS, Nikon/Canon whatever) locks you into old tech with no hope of improvement.

7 upvotes

Until we see a firmware fix, I'm not going to assume that it's easy to do. We simply don't know the extent to which the bit depth reduction, tone curve, and block encoding are baked into the signal processing acceleration ASIC.

4 upvotes

If there is a bandwidth bottleneck, no firmware is going to fix it.

1 upvote

Sony isn't particularly renowned for adding features in firmware updates. They add lenses and tweak a couple settings, but if the feature isn't there when you bought the camera it's not likely to appear in the future. That being said, they have publicly acknowledged the lossy compression issue as something that could be improved with firmware, which is a large step further than they have gone on any other issue they have ever commented on. So perhaps this might be the one new thing they introduce. But if history is the judge, It will appear in the A7SII and beyond.

1 upvote

The fix will be called a7 r iii ;)

1 upvote

And it'll cost $5000.

1 upvote

@Photoworks: Compression is used even in computing power limited environments, because the overall system performance is defined by a "few" more things than raw CPU speed, especially if we talk about systems where you have multiple CPUs, so to speak (ignoring the fact the actual CPU could be multi-core). For example, if you have a bandwidth limited I/O device, compression can actually help increase the total throughput and decrease the overall resource usage. Not to mention that the compression we talk about is kinda straightforward to implement.

Aside of that, I second what exapixel wrote, because we have no idea where this compression actually happens and how much of it is baked into the whole signal processing pipeline, rendering it more difficult to change in any way.

1 upvote

This reminds me of the Canons of a few generations ago when photogs figured out that certain higher ISOs on the 5D's were better than lower ISOs. They had quite a thing going there for a while.

My main question would be what do you tell a client when he or she sees the photos coming off your camera and they're way too dark. 99.9% of people cannot make the exposure adjustment in their mind so you'ld end up with a pretty freaked out client. In the Cinema world we can load look up tables to correct the image before it goes up onto the monitor for everyone to see. maybe there something you could do similarly with these images.

0 upvotes

That's what S-Log2 Picture Profile is for. :)

Or even DRO Lv5, to a certain extent.

1 upvote

BTW on the Canons the Technicolor Cinestyle profile is allegedly S-shaped so I wonder if that would do something similar? It was originally designed for the 5DII and I've only just started playing with it on my 5Dsr.

0 upvotes

Sony's groundbreaking camera is truly innovative. I believe it's the first camera to manifest more artifacts at lower ISOs!

18 upvotes

You can try Canon 5D original for the priority

5 upvotes

You don't know what you are talking about...

4 upvotes

Comment edited 3 minutes after posting

0 upvotes

Great article. I'm very impressed and want to buy this camera. I need to learn more about the native lenses.

0 upvotes

A simple way to see on the a7rII, when you underexposed is to turn OFF Settings Effect, Live View.

Rishi, great article, and useful.

A great point on 14-bit RAW.

Though I would probably use a True LossLESS Compressed RAW more :)

1 upvote

This D810 camera that you keep mentioning seems pretty interesting, and I look forward to reading an in-depth review when it comes out.

(I kid, I kid).

15 upvotes

1 upvote

Meanwhile an Olympus update appears and within hours we get a full review. Go figure.

0 upvotes

The ISO100 was so bad, unreal. Half-cooked raw?

6 upvotes

Which cameras at any ISO would you say look good when pushed 6 EV?

5 upvotes

The question is how to make it possible.

0 upvotes

Those are 11+7 bit RAW artefacts. Something most Sony fanboys say do not exist.

5 upvotes

I know DPR's policy is to use ACR because of its market dominance, but doesn't ACR incorporate undisclosed "optimisations" that vary with the ISO? Because if it does, then the results of this test will be affected, perhaps significantly.

A test of ISO100 vs ISO1600 (4stops) should be performed using both ACR and RawTherapee (an open-source raw converter whose neutral processing behaviour is verifiable). If there is no difference, fine.

I'm not against raw converter optimisations if they can be disabled when required. But if they are present in ACR and can't be disabled, then this is not a test of "true" ISO invariance, but of "real world" behaviour i.e. what ACR has been specifically designed for, with ISO100 shots expected to need less noise optimisation than ISO1600+ shots.

Dan.

Comment edited 12 seconds after posting

3 upvotes

I just had a look at the iso100 raw in RawTherapee and the artefacts are more pronounced.

5 upvotes

WOW. this well lit night street light shot was NOT exposed ETTR for highlights at all @ ISO 100. Whilst ALL fake-shadows in ISO 100 may have been pushed recovered (de-faked; uncooking cooked RAW) in pp, but the real shadows REMAIN very BLACK (unrecoverable), all blown highlights remain blown (store display was normally lit, yet poorly, and grossly OVER-EXPOSED).

this is a perfect example of confusing users of fake high DR using pre-cooked high contrast where the faux blacks are overlayed on top as fake ISOs are 'simulated'.

thus, Sony (and by extension Nikon, others), never really have true exposure simulation (nor real ISOs = ISO-less = ISO Invariance = in-camera ISO 'faking'). they are likely just making do with an actual fixed ISO 1600-3200 (judging by noise) sensor, and simulating ISO (cooking RAW with numerically adjustable shadow overlay = removable in post of course)

a sensor showing true uncooked RAW is by far much flatter looking, and easier to gauge blowouts or shadows

Comment edited 2 minutes after posting

5 upvotes

Please re-read the article, carefully. You've misunderstood what was done in this test.

And please stop with the 'true' vs. 'fake' ISO. If a camera is actually ISO-invariant over a certain range of ISOs, you actually gain dynamic range (on the high end) but saving the amplification for later. It's not detrimental, as you suggest.

Furthermore, manufacturers aren't even programming their cameras to work this way; I'm suggesting you work this way in order to give you more dynamic range. So your claims of manufacturers using 'fixed ISOs' and thereby 'simulating ISOs' and therefore 'cooking' their Raw is completely wrong at best. And ideally, this is actually what manufacturers would do, so your suggestion that it's somehow bad is also misinformed.

27 upvotes

@Rishi I think Sdaniella is a Canon fan boy and/or paid troll and is jealous of Sony's new tech. Maybe he/she believes the world is still flat...

5 upvotes

Sdaniella's posts are often good for a laugh, but this one is pretty special!

4 upvotes

@RPJG Maybe today is special people's day ;)

0 upvotes

to wit:
if a live preview shows you extent of both highlight and shadow details straight from the sensor, why would one intentionally adjust ones ISO and Exposure settings (contrast/gamma curve; speed+aperture) such that highlights are distinctly SEEN LIVE (preview) are NOT BLOWN (@ISO 100) ... yet, the image saved show it blown?! Does this not indicate a STRONG indication that there is a DISCONNECT between what is SEEABLE vs what is RECORDABLE (if data is dropped/tossed/altered; LESS is recorded to save space). The dumped data is 'reconstituted' in post via data adding = push processing 'blacks'.
I.e. if one does not have variable ISO on a sensor, one can mimick various ISO settings in-camera, which is the reverse in post, correct?
do you get my meaning? its mathematical processing going on, not gain changes at all
maybe, this is the only way to achieve smaller RAW file sizes, and why JPEG sizes are only slightly smaller than ARW files (allows higher fps, too)
yes, I read the article

0 upvotes

I don't dispute high end ISOs are 'real', because on the upper end, gain is applied on sensor (no data cooking (no data dropping) is needed).

I'm certain the store display uses ordinary lights much lower in intensity than 'full daylight' which ISO 100 could easily NOT blow with the appropriate shutter speeds!!!

think about that! indoor store subject lighting is a perfect 'middle grey' wrt EV settings, and shouldn't clip (blown as shown).

Comment edited 5 minutes after posting

0 upvotes

Rishi: Were you also involved in those fake moon pictures NASA sent out to make people think we actually had astronauts on the moon?

0 upvotes

imagine:
if all one had was a contrast and gamma sensor control, to ensure high contrast shooting scenarios can dial out film era contrast, so DR is better previewed LIVE, rather than obscured.

one could SEE a live RGB/Luminance histogram get adjusted so it gets narrower, if it is too wide (to prevent dark and bright ends from going beyond clipping on either ends).

u may laff at me 'cuz I don't use engineering tech terms I don't know, but just use plain layperson terms ...

some cameras offer adjustable live contrast (live DR) preview = I think useful.
others, do not. cuz they fix in post.

0 upvotes

Oh boy, yeah I guess this isn't even worth trying...

3 upvotes

@Rishi LOL!

0 upvotes

Well it seems to me that pushing 800 3 stops works well. Pushing 100 is fine if you like cleaning trash. The one thing I would like to know is if you shot a correct exposure of 100 with -3 stops EC then push back 3 in post would artifacts still show up? Even if lunch was free I don't think I would ever push anything +6.

1 upvote

Given the less-than-1 EV noise cost in shadows to shooting ISO 100 and boosting 6EV (vs shooting ISO 6400), and no noise cost in midtones to shooting this way, I'm not sure I'd call ISO 100 + 6EV 'cleaning trash'... except for what lossy Raw compression does, of course.

Do you ever shoot ISO 6400? If so, you're just having your camera do that +6 EV push. Why is it so unacceptable to do that +6EV push in post as opposed to in-camera?

More to the point: why think it's acceptable to do the push in-camera, but not in-post? Seems a bit of an artificial judgement call... :)

3 upvotes

How to do a +6EV push in ACR?

0 upvotes

My questions are :-

Do other manufacturers do lossless compression(which seems a given, since it would be easy to achieve) and is it something about the lossy compression that Sony uses which increases the artefacts. i.e do other manufactures do a better lossy RAW compression?

It is also strange that Sony did not offer non-lossy compression because I assume that RAW data have to go through another processing stage to be compressed. Just removing that processing would in theory produce the desired effect. The only reasons I can think Sony have not done this is it would affect frame rate due to the time it would take to write the larger files to disk and compression is done by a dedicated processor engine which would be difficult to disable. Any ideas why it is not already an option?

0 upvotes

Nikon cameras offer lossless compression that takes advantage of the properties of shot noise to compress the data (since you waste bits overquantizing noise for brighter signals... you don't need to record with single photoelectron precision when your signal is always varying by the square root of the signal).

Nikon's implementation is clever, and shows no noticeable, or even measurable AFAIK, degradation.

See Emil Martinec's excellent explanation here.

We wish Sony would implement something similar.

4 upvotes

I hesitate to correct Rishi, but it's important to note a distinction with a difference here. Nikon's comparable bodies offer truly lossless compression, a visually lossless but (to a mathematician) technically lossy compression that is completey distinct from Sony's and which doesn't suffer its artifacts, and of course a completely uncompressed output file. Nikon also supports setting the bit depth to 12 or 14 bits. So there's six combinations, and then there's also a new "small raw" option in the D810. In practice, I set 14-bit lossless compression, and get every damn bit of my data.

4 upvotes

In re-reading that sentence, I realize that it's pretty unclear; I'll try again.

Modern Nikon high-end bodies have THREE general kinds of compression and TWO selectable bit depths, and you can combine them any way you like. The compression techniques are (a) uncompressed data, (b) losslessly compressed data, and (c) "visually" losslessly compressed data. The visually lossless compression technique applies a tone curve and then the same lossless compression (non-adaptive Huffman coding) as (b).

As an analogy, think of truly lossless compression as being like printing a book in a smaller font, while lossy compression is more like having Reader's Digest delete some of the words. If you can't reconstruct the original data exactly, it's not lossless.

1 upvote

I think that the "compression" happens already in the sensor by design using fast/low precision differential ADC's. This allows to feed the image processor and the many hardware pipelines with compact and handy 128 bit blocks of data. I guess that this saves a significant amount of silicon, (battery)power and bandwidth through the various hardware and software pipelines. I think that changing that requires a complete hard and software redesign needing faster components and more power/bandwidth.

1 upvote

That makes sense @Foroa and if so, just shows that making changes to complex beasts like this is never as easy as "just changing the software".

However another issue maybe that it would require a different RAW format meaning all the RAW converters would need to be changed.

Saying that Sony have said in the past that they were open to the change, so I am assuming that technically it is possible.

0 upvotes

True, exapixel, there are different modes that Nikon offers, I didn't feel like listing them all out, so thank you for doing so :)

And just to clarify further: Nikon has the following options on the D810:

-Lossless compressed -Compressed -Uncompressed

I assume what you're saying that their 'compressed' option is 'visually loslessly compressed data'?

Thanks for the info.

0 upvotes

Yes. Uncompressed = simple pixel array, lossless compressed = Huffman encoding, "compressed" = tone curve + Huffman.

0 upvotes

Ah, so lossless compressed doesn't even use the gamma curve that takes advantage of the principle of shot noise? If the only difference with 'compressed' is that it uses that shot noise principle, then, yes makes sense to call it 'visually lossless'.

To me, it's also 'practically lossless', since there's really no benefit to overquantizing shot noise... has anyone done a thorough study to show if compressed ever leads to a degradation in image quality? My guess is it doesn't.

1 upvote

Rishi, that's right, and I recommend 14-bit "compressed" when people ask for advice... but when I'm in a once-in-a-lifetime travel destination, I go for 14-bit lossless, because disk space is cheap and travel is not.

Also, Nikon's tone curve is not Sony's, although it's kind of the same idea (quantize to shave off the noisy bits). Sony's tone curve is never 1:1, even in the darkest shadows.

Or put another way: Sony's "14-bit" raw only has 2048 - 255 = 1793 possible distinct values per pixel channel, due to black level subtraction and 11-bit coding.

1 upvote

My first DSLR was a D7000 and I've been using similar techniques since I had it - I shot a lot of architecture at night off a tripod.

One technique I used for composition and focus is to to use LV and jack the ISO up, punch in if necessary, then reset ISO to base and shoot.

In some conditions you may need to open the aperture too.

0 upvotes

Noob question - what about shooting lower ISO (by 1 stop) and pushing later to preserve fine details (like critter fur or feathers), rather than highlights? Would that help improve detail retention or is that just silly?

0 upvotes

Few points:
1. Looked at 100% all ISO samples are crappy. There is no free lunch yet.
2. As far as the noise and detail goes every ISO is better than lower ISOs pushed for the same brightness. The difference between ISO100 and 6400 is not modest, it is huge. So, your recommendation should be to use the highest ISO (up to ISO6400 for this test) in low light without blowing the highlight and push as less as possible in the post. This is quite opposite to the tone of your article or recommendation. Moreover, higher ISO let one use higher shutter speed in low light handheld situation
3. We don't have a 'true' ISO-invariant camera yet. All modern cameras become ISO invariant at certain ISO.
4. Whenever possible one should aim for proper exposure in the first place. Pushing shadows comes with cost no matter which camera you are using.
5. Best image in this article is the one from D810.

4 upvotes

Yes, it's stupid to push ISO 100 files to 6400 equivalent brightness. But surely you can see that 800-6400 is quite similar, and 100-400 is another band of similarity? Jim Kasson had demonstrated this with data a few weeks back. DPR were wrong to call it a blanket ISO invariance - there's 2 bands of it, but they're darned good. For a wedding photographer, I know if I had an A7rII shooting low light I could stick to ISO 800 and always get the shot I want regardless of lighting condition (i.e. manual flash setting means ISO 800 always), which is great to know. And in daylight, shoot at ISO 100 and ignore all else.

4 upvotes

> it's stupid to push ISO 100 files to 6400 equivalent brightness

And what do you think ISO setting is doing? Just that, pushing the sensel output.

Not even discussing there is no brightness until the image is not displayed or printed...

Comment edited 1 minute after posting

2 upvotes

I'm talking about pure +EV push in PP. If that was all the ISO setting did, then you'd already have ISO invariant cameras.

2 upvotes

PP or the camera, and the difference is?

Maybe I missed somehow, and the word "stupid" changed its meaning and now is considered to be not offensive?

4 upvotes

The A7rii is iso invariant form ISO640 to 25000, as Jim Kasson showed (the sensor changing the conversion gain at ISO 640). So it is probably better not to start form ISO100 on. This is different in the D810 and A7r, for example. So if you want to preserve highlight, one should keep that in mind. In that case, a proper exposure for me is the one, where I ETTR and push in post.

3 upvotes

Please educate yourself here IB: http://blog.kasson.com/?p=11135

The conversion gain is likely different from ISO 500 to 640, meaning that ISO 640 files are actually cleaner than 500. If you did that with an EV push, they would be noisier. So the moral of the story is that if you are going to ISO 6400 equivalent push, start from ISO 800 (or actually 640), not from ISO 100.

As for the "stupid" remark, after seeing the evidence posted by DPR above, wouldn't you agree that it's stupid to push an ISO 100 file to 6400? Much better to start from ISO 800 right?

Comment edited 3 minutes after posting

3 upvotes

So, exactly why it is stupid to test to see the difference, to present the results, and to analyze why it happens? It is a good community service. Seems you are missing the point.
It is you who may need education here, and not only in digital imaging.

1 upvote

@HFLM : to do a test, one may start from the lowest. That is how we know where to start for production work, right?
Another matter is that extremely high contrast edges are not present in every scene, and in my testing the lowest ISO on SONY cameras was never the cleanest, so in a lot of cases ISO 200 or ISO 320 setting is quite OK, and does not clip highlights as much as 640 or 800; while exposing -1 EV for ISO 800 to have the same clipping point as with ISO 400 results in more noise.

1 upvote

@aftab: Sorry for the derail. Anyway, DPR's second footnote (**) explains well the issue:

"Bill Claff's data on the a7R II vs the a7R indicate that at ISO 640 and above, a higher analog gain is applied fairly upstream in the signal pathway, which can help overcome the finite downstream read noise that, while incredibly low in Sony sensors, still is non-zero. In this case, then, 'ISO-variance' is actually a good thing, with the camera only truly ISO invariant from ISO 640 and above."

...Though it's not exactly accurate, as there's a second boost at ISO 32,000, as the Kasson article explains.

Comment edited 2 minutes after posting

0 upvotes

> Weird, the ignore function doesn't work in article comments.

And why did you need to tell me that? "Manual" ignore mode does not work, too?

On all SONY cameras I used the lowest ISO setting was not the best for deep shadows, same as it is with many Canon dSLRs. It changes a little with pixels being smaller and well depths getting smaller as the result, but still it is something to test and consider. On some cameras there was no actual reduction of dynamic range when going 1 to 1 1/3 of a stop higher than the base ISO, just a shift - the loss in highlights was compensated by the gain in shadows.

As to the benefit of "higher analog gain", that is exactly what this "stupid" test is for - to see the results in a torture test and adjust the shooting style accordingly.

I gather quite some of the SONY shooters missed the Jim's and Bill's pages, but may look at this "stupid" test and benefit from it.

0 upvotes

Celsus, I agree that difference between ISO800 pushed and ISO6400 in not that great in most part of the scene in this test. But in some places it is quite visible.

http://www.dpreview.com/galleries/4269542963/photos/3277684/2015-08-25_114429?inalbum=this-and-that

Wish we had ISO3200 to compare.

0 upvotes

@ aftab : 3200 is available in a drop-down.

0 upvotes

Thanks Iliah. Being a DPR-holic for so long I should have thought about it. Anyways, barring the blown highlight ISO3200 pushed and ISO6400 seem virtually indistinguishable, although I think in the deepest shadows ISO6400 wins by a whisker.

0 upvotes

@ aftab : I think I could use ISO 2000 and careful conversion. An interesting thing however is to take highlights from ISO 200 shot exposed for highlights and shadows from ISO 1600 shot but exposed for shadows (or something along those lines).

0 upvotes

@Iliah: I guess ISO1600 to 3200 would be reasonable in most situations. One more thing I have noticed is that ISO6400 seems to be the brightest of the lot in this test. It could be due to the RAW converter.

0 upvotes

@aftab. We are talking about difference at 100% view! Do you really print ISO 6400 shots at 100% view? Apply some NR and ISO 800 or ISO 6400 printed to A3 shouldn't make any visual difference.

0 upvotes

Celsus: whoa buddy, you really, really don't tell Iliah Borg to educate himself, and if you do, you're only showing your own ignorance :)

Iliah is spot on in everything he says. You are also correct in most of what you say, save for the following:

It wasn't a 'blanket ISO-invariance' we grouped everything under, since our footnote explained the more nuanced explanation of dual conversion gains. It's just that if that's in the meat of the text, things get more complicated than they already are, which doesn't exactly help our readership... I put it in the footnote, and said we can discuss further in comments. Seemed like a reasonable thing to do.

Furthermore, you say: "As for the "stupid" remark, after seeing the evidence posted by DPR above, wouldn't you agree that it's stupid to push an ISO 100 file to 6400? Much better to start from ISO 800 right?"

No, not stupid if you need the extra highlight headroom. The noise cost in the shadows isn't exactly life-threatening at ~1/2 EV.

0 upvotes

The second gain at ISO 32,000, btw, is likely noise reduction baked into Raw.

Just a more general point though, which I've added to the article: at worst, the noise cost in the shadows of an ISO 100 + 6EV shot vs ISO 6400 appears to be slightly less than 1EV, and in midtones the noise cost is almost nil. But on the upper end, the gain in highlights is huge.

Therefore, while I agree that you should 'start' from ISO 800 if you're aiming for an ISO 6400 exposure, you don't have to, especially if that still doesn't save your highlights. In that case, I'd go to a lower ISO, and pay the at most 1 EV cost in the shadows (but nowhere else), b/c I just gained 1 EV in my highlights. That'sw why DR always goes up with lower ISO (save for at that conversion gain step), and at worst you just pay a small noise cost in the deep shadows. Yes I understand your point about the different bands of ISO invariance, but in the real world, slipping from one band to the other won't ruin your images.

0 upvotes

Nasty blown highlights, OTOH, can ruin your images, IMHO. Which is why I like the freedom of being able to err on the left, even when I'm technically trying to ETTR. Especially since we have no Raw histograms, so I don't even actually know when my Raw is clipping. Better to be safe, which is why with the D810 horse shot, e.g., I went down to ISO 140 without too much concern, despite my knowing that I'm paying some quantization noise cost there.

Which brings me to another point: ISO 100-400 is not entirely ISO invariant, b/c quantization noise goes down as you go from 100 to 400, since at ISO 100 a 14-bit ADC isn't enough to adequately describe the full well, and so every photoelectron can't be counted. So you lose counting precision, which you gain back by going to ISO 400, or wherever 1 photoelectron is counted as 1 digital number (DN). Unless there's some interesting dither or nonlinear behavior in the sensor --> ADC mapping, which would be news to me?

0 upvotes

> composition you are shooting needs all 14 stops of dynamic range

But dynamic range one can really achieve and use is no more than 11 stops usually.

Even if the shutter button is pressed twice.

Here is why: http://www.rawdigger.com/Dynamic-range-fair-share-of-flare-and-glare

0 upvotes

I'm removing all comments with personal attacks. We don't wish valuable members of our site to feel offended on, well, our site.

Also, your defense sounds a lot like dynamic range apologists from you-know-who-brand owners. It would be disingenuous of us to point out dynamic range shortcomings in those cameras, but not talk about them in Sony cameras.

Also, the ironic reality is that in "real-world" shooting scenarios, I'm not perfectly optimizing my ISO. I literally can't, b/c manufacturers give me no love with Raw histograms. So I often purposefully keep the ISO low to make sure I don't blow out, say, the moon in my shot.

And then when I post-process that file, I get ugly artifacts around the moon. I don't with my D810. That's not cool, to me, as a photographer, nor as a geek. And it's worth reporting on, in my honest opinion. I'll publish this shot I speak of shortly, btw.

1 upvote

So I am wrong to think that bracketing catches different parts of the dynamic range of the scene better than a single file pushed all over the place?

0 upvotes

My final post: I have not once said anything about what the manufacturer should do. I have only been discussing what the shooter should do to achieve ideal output given existing constraints. I don't know how I ended up being so completely misconstrued but it's left a very bad impression about this site.

1 upvote

Total comments: 369

Editorial content

Product database

Community

About us Work for us Advertise with us FAQ Feedback Interest-Based Ads Privacy Legal

All content, design, and layout are Copyright © 1998 - 2015 Digital Photography Review All Rights Reserved. Reproduction in whole or part in any form or medium without specific written permission is prohibited.

Mobile site


Sony Alpha 7R II can match or beat DSLR low light AF performance Aug 20, 2015	The new Canon 35mm F1.4L II will be a thing of beauty Aug 28, 2015	BSI Boss? Sony Alpha 7R II added to studio scene Sep 1, 2015	Readers' Showcase: Tony Eckersley Aug 30, 2015

98	* This Week Through YOUR m4/3 2015.09.05 *
75	Dynamic Range
66	12/2 or 15/1.7, I can't decide...
62	Mirrorless vs DSLRS
52	A sobering thought
49	Who bothers with a lens hood on m43 these days?
44	Does anyone else have problems selecting the focus spot?
38	Message to Panasonic: Thanks for V-Log, but I'm OUT!
38	Do you think your photos are better because you shoot raw?
37	The lossy compression artifacts challenge
36	Your thoughts on FX-DX-CX system combo
35	Fuji Canada refusing to fix 9-month old X100T - any recourse?
35	Is sharpening an image (RAW) useful?
34	Sony A7xx cameras vs Canon 6D high ISO performance
33	Who is Collecting Old Digi Cams?
33	Why we go to "details" and "sharpness" instead "moment" and "possibility"?
32	GX8 any problems/
31	** Weekly Wildlife Thread 9-5 to 9-10-2015 **
29	Blu Ray successor?
28	Bring back the Nikonos (or rather, dedicated aquatic cameras).