a7S high ISO claim - - -over hyped or not ?

Started Jul 3, 2014 | Discussions thread
Shop cameras & lenses ▾
Rishi Sanyal
Rishi Sanyal dpreview Admin • Posts: 607
Re: a7S high ISO claim - - -over hyped or not ?

spacemn wrote:

Rishi Sanyal wrote:

Well, I think the bigger issue here is that physics is physics, and once you've already gotten read noise down as low as many of these cameras have (through sophisticated techniques like correlated double sampling), the only way to drastically increase ISO performance is to capture more light. Capturing more light increases SNR by decreasing shot/photon noise contributions.

And the only way to do that is to (1) increase sensor size, or (2) increase sensor efficiency. Or, of course, use faster glass.

So Sony's investment in MF probably reflects these realities, to a certain extent. Increasing sensor efficiency is difficult, as far as I understand. Partly b/c sensors are already so good, and also b/c it's difficult to mitigate perhaps the largest source of light loss - the color filters. Short of making a 3-CCD/CMOS with 3 different light paths, that is.

By the way, the higher SNR due to lower shot noise I referred to above is also what gives medium format 'that medium format look' (aside from conversations about DOF and bokeh, of course). It's the same reason a shot taken at ISO 100 looks better than the same shot taken at ISO 400 on even the best performing camera. The ISO 100 shot received 4x as much light as the ISO 400 shot (assuming you matched exposures accordingly); since bright shot-noise limited tones have SNR = sqrt(signal), the ISO 100 image would have twice the SNR of the ISO 400 shot in well-exposed regions (so we're ignoring read noise and tones of the image affected by read noise).

Sony's new MF CMOS sensor has 1.68x the surface area of full-frame, so theoretically can potentially give you almost 2/3 EV ISO performance advantage. But that's assuming similar aggregate read noise performance (and remember, with 50 million pixels - that's twice as many pixels to be read as a 25 MP sensor), similar QE, and of course similar access to fast lenses (which is definitely *not* the case). And, by the way, before this MF CMOS sensor from Sony, previous MF digital sensors had really, really low performance (either low QE, high read noise, or both) that limited their ISO performance despite their theoretical advantage due to their size. But for well-exposed images (landscapes, studio lighting, etc.), you couldn't beat those MF shots (well, unless you used multi-imaging techniques). Not just because of their resolution, but b/c of the very high SNR due to lower photon/shot noise contributions.

And while we're on this subject - you can increase the SNR any camera is capable of by averaging multiple shots. For example, take two APS-C shots, align them, then average them together and you'll basically get pretty close to what a FF camera might have gotten given the same focal plane exposure. This is simply b/c averaging two shots is like collecting twice the amount of light as far as photon statistics is concerned - and a FF has ~2.25x the surface area of APS-C and so, generally, collects twice as much light given the same shutter speed/aperture. Neat, huh? It doesn't perfectly work out b/c the two APS-C shots together had twice as many pixels read as the FF would have (assuming equivalent resolutions). But, generally speaking, SNR increases as the square root of the # of images averaged.

Anyway, to answer your question: I think it is somewhat fair to say that sensors have gotten so good that we're running up against practical limitations that'll require quite some innovation to overcome. But look at how far we've come since the advent of digital cameras. Who knows what's on the horizon?

-- hide signature --

“It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts.” -- Sherlock Holmes

I am not sure why you are bringing in MF sensors into this discussion. The first MF sensor which seems to be outperforming the FF sensor when it comes to SNR or high ISO performance is the new Sony CMOS MF sensor. Before this there were only CCD MF sensors and their low light performance was appalling in comparison to modern CMOS FF sensors.

The only place where MF CCD sensors have excelled over FF CMOS sensors have been colour bit depth and tonal range, and then of course bucket loads of resolution especially the 60-80 mpix sensors.

So, can you show us this "MF look" when it comes to SNR where it beats the best FF CMOS sensors? to my knowledge only the new Sony MF CMOS sensor seems to have handsomely beaten the FF counterparts and hence introduced a new "MF look" when it comes to SNRl, but this sensor is new, and the MF scene is still dominated by way inferior lowlight CCD sensors.

The performance of the sensor is comprised of many factors. Such as chip technology and the signal processing, and not at all only the size of the pixels. It is only matter of time until we will see a new chip tech generation coming along. Perhaps next disruption will be based on organic light sensitive diodes.

No offence, but if you are in any case involved in the A7S review on behalf of DPreview, then I really hope you have your facts under control and won't make fluffy assumptions such as "MF look" etc.

To thread starter, no the A7S low light performance is not overhyped in my opinion. Especially at super high ISOs (>6400) it is significantly better than anything else on the market (we are talking <2 stops better than most popular FF sensors on the market), a little revolution I'd say. Photographers and film makers can now use candle lights and moon light as creative lighting now, which was not really realistic earlier. DXOMark scores indicates this as well. Maybe only the new Sony CMOS MF (medium format) sensor can beat it. It seems it both focusses and performs like a champ in almost total darkness

Perhaps you missed my sentence: "And, by the way, before this MF CMOS sensor from Sony, previous MF digital sensors had really, really low performance (either low QE, high read noise, or both) that limited their ISO performance despite their theoretical advantage due to their size."

I am quite aware of the poor performance of previous generation MF sensors. Furthermore, the reason I brought up MF sensors is b/c I was responding to a previous comment that specifically brought up Sony's new MF sensor, asking if we'd reached some sort of limit in terms of noise performance on FF. A bigger sensor is one way to (theoretically) address noise performance.

As for 'fluffy assumptions', my entire point of bringing up the 'MF look' was to debunk this fluffy assumption to begin with. There's nothing magical about the 'MF look'. The clean 'MF look' results from the same principles by which an ISO 100 image looks better than an ISO 800 image on the same camera.

Your question about MF SNR beating FF CMOS: the SNR increase is quantifiable, though whether or not you *see* it is another matter. B/c for well exposed tones, the SNR is already so high that sometimes it's difficult to appreciate even higher SNR. But even in a well-exposed image, you may have tones (e.g. dark blue skies) that will be less exposed and therefore have lower SNR. For such tones, capturing more light by having more sensor surface area can help, b/c you'll have lower overall photon/shot noise. Unless, of course, read noise is so poor as to have an effect (probably not, for well-exposed tones).

My overall point was: MF sensors have the *potential* to outperform FF, just as FF has the potential to outperform APS-C, and APS-C has the potential to outperform mFT. Whether each format does or not depends on sensor performance & attached lens (when you're light-limited). When you're not light-limited (e.g. landscape photographers using tripods), all else equal, whichever sensor has a higher total full-well capacity (FWC of pixels x number of pixels) will win, for tones well above the sensor noise floor.

By the way, re: higher bit-depth ADCs - unless you have a high DR sensor, higher bit-depths don't necessarily help, & often just over-quantize noise. This has been shown no better than by Emil Martinec in his fantastic treatise here.

Not sure why you fear me having my facts under control when you clearly didn't even appreciate the major point of my entire discussion. If there area any particular 'facts' that you take issue with, please enumerate them & let's have an intelligent discussion.

Also, we've done an extensive study of A7S low light performance, and will be adding more cameras to the comparison shortly. We'll then correlate them with full SNR curves, so that there's really nothing open to debate. In case you haven't seen it, here it is.


 Rishi Sanyal's gear list:Rishi Sanyal's gear list
Sony RX100 IV Nikon D810 Canon MP-E 65mm f/2.5 1-5x Macro Nikon AF-S Nikkor 85mm f/1.8G Sigma 35mm F1.4 DG HSM Art +3 more
Post (hide subjects) Posted by
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow