Strong n2 noise means that the camera can still have very good high ISO performance, but it will be limited in DR in lower ISOs - and you can't "push" the file very much in post before you start pushing the n2 into visibility. This is the Canon sensors, the Nikon D3/D4 sensors, and the older Samsung sensors.
Very low n2 noise figures in sensors like the Exmor, the newer Panasonics, Aptina, Samsung sensors means they have good base ISO DR, but they're not in any way necessarily better in high ISO performance (than the "strong n2" sensors).
There are several more parameters too of course, but actually those two - n1 and n2 - are the main performance determining factors next to color filters.
mco_970 wrote:
This ^. Some color channels can blow out with the histogram looking just fine (reds and yellows, ugh) and so if anything I am more likely to ETTL to preserve colors.
However, I shot 4/3 for a while a few years back, and ETTR made a huge difference then. Seems like the camera bodies then tended to underexpose, and were unforgiving for lifting shadows. It was less of a big deal when I moved to Canon, and now is mostly a non-issue with Exmoor sensors.
Now days, I do ETTR a bit with Sony if I'm in really low light (high ISO), and I know highlights are going to make the camera underexpose. Then I'd rather blow out a few highlights and have less noise in the overall image. I feel like there is only so much you can pull up exposure at ISO3200 and beyond w/o IQ becoming friable....Show more →
Exactly. I think most of this ETTR theory came about in terms of indoors high ISO/low light situations - wedding photography, etc. But in terms low ISO landscape or the type of general outdoor photography that I enjoy ETTL works much better for me.
RustyBug wrote:
I'd mostly heard that large pixels = low noise for low iso, small pixels = low noise for high iso (or some corollary along those lines) ... never really sure-sure if I understood it correctly or why? I shoot mostly base ISO 160 with large pixels on SLR/C, and occasionally shooting some ISO 3200 with 1D II N for low light (also shoot 1D II N @ base ISO 100 - ISO 400 for "action" @ sufficient light).
Given those two cameras ... and maybe something like the Pentax K3 (looked at today), how do we know when to better harness n1 vs. n1 + n2? ...Show more →
I wouldn't want to simplify large/small pixels like that any more, it all depends on your usage and what you do with the files. And who that manufactured those "pixels"...
In general, "good" or "enough" resolution is about twice as many MP in the camera as you need for the end result image. Smaller pixels than that WILL cause losses due to manufacturing limitations and so on at higher ISOs. At lower ISOs, there's actually almost the inverse relationship - more is better.
Remember that both n1 and n2 are synthetic noise sources, electronic value errors added ON TOP of the original light source noise (shot noise). The more of them you have, the more you degrade the image. But smaller pixels often have lower per-pixel n1/n2 values...
We can now make small pixels so good that you can't see much difference (in per image noise performance) between the D3s and the D800 at ISO3200-6400. That's even though the D800 adds in 3 times as many electronic error values to the image data...
But at ISO25600, the difference is obvious. The D3s is much cleaner, and it actually gives a higher usable end result image resolution (than the D800). That's because of the number of noise sources added in - even if the error value added by each pixel instance in the D800 is lower than those in the D3s, there's so many more of them. And extremely high ISOs are limited by per-pixel electronic noise in a totally different way than low ISOs.
Love the K3 btw... But it's not for me Excellent camera, love handling the latest Pentax (now Ricoh...) bodies - but I never really "got" their lens line-up.
RustyBug wrote:
Judging from your tests (both Phillip & sebboh) ... which of the two exposure strategies would you suggest using as best for highlights, best for shadows?
I hope you don't mind (just say and I'll pull), but here's a rough to see how far I might take the file from the adobe default image ... and the penalties incurred. Highly subjective @ rendering liberties of course.
the best strategy is to bias exposure for the part of the image that's most important to you.
how much better is the RX1's DR than the Nex-7/5n's in your eyes?
Where are we asking? at ISO 100 I haven't noticed a difference between the A99 and the Nex-7 maybe a bit more headroom with the A99 sensor but I've only seen that in severely over exposed ISO50 shots where the A99 was able to recover 3 stops.
At ISO 3200 I never felt I had any more than about 1/3 of a stop of latitude with the Nex-7, I will push 1.5 stops nonchalantly with the A99.
Keep in mind that my shots are mostly indoor candids where my biggest concern is keeping skin tones nice and preserving facial detail. I quite often like to have shadows that actually slightly clip into black.
Where are we asking? at ISO 100 I haven't noticed a difference between the A99 and the Nex-7 maybe a bit more headroom with the A99 sensor but I've only seen that in severely over exposed ISO50 shots where the A99 was able to recover 3 stops.
At ISO 3200 I never felt I had any more than about 1/3 of a stop of latitude with the Nex-7, I will push 1.5 stops nonchalantly with the A99.
Keep in mind that my shots are mostly indoor candids where my biggest concern is keeping skin tones nice and preserving facial detail. I quite often like to have shadows that actually slightly clip into black....Show more →
I am a landscape shooter, so base ISO for me.
So you think there isn't much more detail in the shadows but some more room in the hightlights?
It seems that way (more highlights) sadly I no longer have a nex7 lying around or I could do a test. Its probably worth waiting to hear if anyone else has experience with the two systems who does more shadow boosting. I never felt DR limited with either camera at base ISO, though as far as I know the nex7 has a different CFA than the nex6/5n so my experience is only applicable with respect to the nex7.
Phillip Reeve wrote:
how much better is the RX1's DR than the Nex-7/5n's in your eyes?
don't really know how to measure that. luminance noise wise pulling the shadows up it is only a tiny bit better (but the extra pixels help there versus the 5N for final pictures), but the color and tones seems to be preserved much better and it's less susceptible to magenta tint – maybe a stop better in that way. same basic story on highlights, i get maybe a 1/3 stop more before the highlights fully clip but it seems to be much less susceptible to partial clipping of color channels and better looking transitions to clipped areas.
at high iso i find it to be about 1.5 stops better, but i think part of that is the lens. usually when i shoot at high iso with the NEX i'll be using an f/1.2 lens at full aperture, which has much lower contrast than the rx1's lens at f/2.
carstenw wrote:
Phillip, that may only be one stop in some technical way of measuring it, but just looking at it, it is horrendous. Whereas the Sony has nice crisp, monochrome noise, the Canon has colour sprinkles, banding and blotching. One stop will nowhere near fix that, visually, IMO. I would be surprised if two stop would.
Canon really needs to bite the bullet and get a new fab up and running.
I think this is a correct evaluation. There may be X amount of stops of DR but how many of them are useable is far more relevant to real world shooting. I had to do a 4 stops pull on an images shadows recently (got caught shooting something that needed fill flash but I didn't have one on me, don't ask) the shadows were appalling on my 5D3. The difference between highlights and shadows in the scene was 7 or so stops as metered. If I believed DXO then I should have been able to have that image without a problem when exposing the highlights as far right as they would go without blowing. Either that or I don't understand what they mean when they say a camera has X stops worth of DR.
Been shooting a lot of commercial work recently which has necessitated annoying bracketing and HDR-ing to hold the scene. To the extent that the A7/r was appealing more for the DR for commercial work than anything else, I just know that higher contrast scenes cannot be shot with a single frame on my 5D3.
I thought that the ML hack, dual iso, could help but the aliasing is unusable. I've tried using their Auto ETTR mode. Sounds good in theory but the moment you get a light source in the frame it's unusable. You can't really automate what highlights a person wants to ignore.
Anyone know if capture one is any better with canon shadows? We use it in the studio, perhaps I'll switch to it if it's better. I prefer the Adobe workflow and funnily enough, colour, but really trying not to have to switch systems.
sebboh wrote:
the best strategy is to bias exposure for the part of the image that's most important to you.
That works in kind of strange ways, when you look at the practical implications...
1- If the "important bit" is darker than scene average, you're going to blow quite a lot of highlights in the rest of the scene
2- If the "important bit" is lighter than scene average, you're quite probably going to induce more noise than necessary
Version "2" here is the case where ETTR works, and gives you better end result quality. You're better off pulling the image in post - that induces less shadow noise than exposing for the detail or indeed underexposing and pushing.
In version "1", you have to chose how much highlight clipping you can accept.
And if the detail matches the scene (no Ev compensation needed) - no difference...
ETTR is really only valid in scenes where you don't have highlights that extend way past the normal exposure. Low DR, or "flat" scenes.
The higher the scene DR is, the more you need to consciously take a decision on what to sacrifice.
theSuede wrote:
You can simplify the noise additions to the signal in the camera into two stages - one is close to the pixel, one is close to the A/D conversion. The "real" analog ISO amplifier is closer to the pixel. I'll call them "n1" and "n2" here. NB - this is not a normal convention.
In low light, you need to amplify the signal to make it brighter - the question is just where you do it.
If you do it in post, you're basically applying a multiplication to the SUM of n1 and n2.
If you do it in the analog ISO amplifier close to the pixel, you're just multiplying n1 - the late stage noise n2 is the same as before, and is added as normal without multiplication!
Say we need a 10x brightening (go to ISO1000, if the camera's base ISO is 100). I'll throw in some arbitrary numbers of n1=5 and n2=20 just to show what happens.
Brighten in post:
Total noise = (brightening factor) * (sqrt(n1^2+n2^2))
>>>> 10 * sqrt( 5^2 + 20^2 ) = 206.1
Brighten in analog, with ISO amplifier:
Total noise = sqrt ( (brightening factor * n1)^2 + n2^2)
>>>> sqrt( (10*5)^2 + 20^2) = 53.9
-the noise is much lower when you use the on-sensor ISO amplifiers, because of the fact that you don't have to multiply the late stage noise too...
In cameras with very low n2 noise (late stage noise) you don't lose as much by amplifying in post. Say that we lower n2 to "10" in stead - that's like getting quieter A/D converters and a cleaner signal path.
Post : 10 * sqrt( 5^2 + 10^2 ) = 111.8
ISO : sqrt( (10*5)^2 + 10^2) = 50.9
Note that the "push in post" noise got a LOT lower - but the "amplify with analog ISO" didn't get much better at all. That's because of the n1 having to be amplified no matter what, there's no getting around that if the image is to get brighter.
That is (fairly simplified) why you can push a Sony sensor quite high from base ISO without the image falling apart, but you can't do it with a Canon sensor. The Canon sensor has good high ISO in camera, since it has excellent n1 noise. But it's limited at low ISOs by the rather strong n2 noise, and that's amplified by pushing in post.
This is also why Canon sensors have lower base ISO DR's, since DR is dependent on both n1 and n2. Higher ISOs are almost purely dependent on n1....Show more →
Here's an example showing exactly that effect, with a Canon sensor:
Beni wrote:
If I believed DXO then I should have been able to have that image without a problem when exposing the highlights as far right as they would go without blowing. Either that or I don't understand what they mean when they say a camera has X stops worth of DR.
DXO define the DR as the difference between the saturation point and the point at which the SNR (signal to noise ratio) drops to 1. That last bit is the important part, it means the noise is the same magnitude as the signal, which to most people is totally unacceptable. If they had instead used a SNR criterion of 2 or 3, the results would probably be closer to what most people "feel" their equipment is capable of.
With DXO it's always important to read exactly how they measure things.
Beni wrote:
If I believed DXO then I should have been able to have that image without a problem when exposing the highlights as far right as they would go without blowing. Either that or I don't understand what they mean when they say a camera has X stops worth of DR.
Hi Beni,
I struggled with this too. I've found this thread over at LuLa to be helpful, for as far as I can understand it, that is
theSuede wrote:
That works in kind of strange ways, when you look at the practical implications...
1- If the "important bit" is darker than scene average, you're going to blow quite a lot of highlights in the rest of the scene
2- If the "important bit" is lighter than scene average, you're quite probably going to induce more noise than necessary
sorry, i didn't mean expose for the part that's important, i just meant bias your exposure a little more one way or another depending how much you value the highlights versus the shadows. it doesn't take long to get a feel for what your camera will lose and what it will keep in given type of scene. with sony's latest sensors i find that i rarely need to give up any highlights in order to keep the detail in the shadows that i need at base iso.
Great thread! As a quick experiment in the "ISO-less sensor" subject recently, I shot this on the NEX F3 very dark at base ISO200 (1/30s, f2 on a FD 30/2 manual lens), and raised the brightness in RAW (can't remember how many stops), with perfectly preserved highlights which might have blown at the camera's suggested high ISO "exposure."
Just did some playing. Using ETTR on a scene for a given exposure gave me an iso 1600. Exposing it as I would normally gave me iso 5000. What's that? 1.5 stops? Boosting the 1600 shot to match the 'correct' 5000 shot made sure that in comparison, the 5000 shot had less noise and zero blotchiness. Even if I lost some highlights. Not quite sure what this tells me other than the fact that sensor gain is far better than software gain. Oh and that if you want to hold your highlights, the shadows are going to hurt as a result. Especially on a canon. Bracket or get a sony sensor.
theSuede wrote:
The higher the scene DR is, the more you need to consciously take a decision on what to sacrifice.
ML also has a RAW histo and Zebra based on RAW data so one can tell exactly what is sacrificed. Auto ETTR is suppossed to allow allowing x % pixels to blow, but that might be a difficult call a prior . RAW Hisot and Zebra would seem to be very useful tools and I think the only other cam with this type of histo has a red dot.
I got nervous when just about to load ML as it has some odd boot flag as auto ETTR and Raw histo are (haven't checked in a month or two) not in a stable build and the boot flag could be difficult to remove. I didn't want to always boot to ML and I don't know how to code to remove the boot flag. I think the latest stable release of ML will remove it though and might be a work around for the non-coding mortals.
