Home · Register · Join Upload & Sell

Moderated by: Fred Miranda

New fredmiranda.com Mobile Site

New Feature: SMS Notification alert

New Feature: Buy & Sell Watchlist

FM Forums | Leica & Alternative Gear | Join Upload & Sell

end

Archive 2013 · My X-E1 vs D7100 High ISO comps

alundeb
Online
• • • • •
Upload & Sell: Off

p.2 #1 · p.2 #1 · My X-E1 vs D7100 High ISO comps

theSuede wrote:
X-trans has very good DR in a downsampled image. It has worse DR "per pixel" in a full-size image than any given Exmor of the same generation.

Bayer based images have the highest available resolution physically possible in a one-layer full-color image. This also means they have the highest DR in a raw-converted image (not necessarily in each raw pixel value before conversion though!).

X-trans has higher accuracy of value in a 3x3-size average cluster, better than if you average a normal Bayer in 3x3 clusters. This means that DR[X-trans] / DR[Bayer] goes up as noise increases.

Compare a high-ISO (downsampled) 1800x1200px image...Show more →

Does this advantage (higher accuracy of value in X-Trans) come from the greater share of green filtered photosites, and the anticipated spectrum of the light matching that, or am I missing something?
Just trying to get an understanding of the magnitude of the theoretical advantage of the high ISO stability.

In a 6x6 matrix, the color ratios are [X-Trans/Bayer] R:8/9, G: 20/18, B:8/9.
That means 11% higher ratio of green light. If I pick a random factor for the spectral utilization of this at say 75%, we get 8% more light with X-trans. That corresponds to 0.1 EV improvement. If we take into account the reduction in color resolution ( as demonstrated as color bleeding in these samples), is there practically anything left?

Jul 01, 2013 at 02:00 AM

FlyPenFly
Offline
• • • • •
Upload & Sell: Off

p.2 #2 · p.2 #2 · My X-E1 vs D7100 High ISO comps

To me, the D7100 has a nicer cleaner ISO3200 but the X-E1 has a much nicer image for when the file is pushed 4 stops. The horrible banding on the D7100 looks terrible.

The WB on the D7100 files seems a bit odd.

Jul 01, 2013 at 06:28 AM

theSuede
Offline
• • • •
Upload & Sell: Off

p.2 #3 · p.2 #3 · My X-E1 vs D7100 High ISO comps

alundeb wrote:
Does this advantage (higher accuracy of value in X-Trans) come from the greater share of green filtered photosites, and the anticipated spectrum of the light matching that, or am I missing something?
Just trying to get an understanding of the magnitude of the theoretical advantage of the high ISO stability.

In a 6x6 matrix, the color ratios are [X-Trans/Bayer] R:8/9, G: 20/18, B:8/9.
That means 11% higher ratio of green light. If I pick a random factor for the spectral utilization of this at say 75%, we get 8% more light with X-trans. That corresponds to 0.1 EV improvement. If we take into...Show more →

Nah, the difference is really not in "average green coverage", it is in "intended color spatial resolution".

Since the spatial resolution of color information is quite a bit lower in X-trans, the optimal raw converter has less guesswork to do. There just isn't any color detail there, so why bother trying to extract it? This means that you can work with better, more stable luminance averages (although also at a lower resolution of course). The opposite of this is Adobes try at the X-trans, that tries to extract a lot of information - and gets it wrong in most cases, creating patterns of strange shapes.

The bane of Bayer is really that noise in red and blue transmits to luminance errors when the missing channel information is built. And also the other way around of course, noise errors in luminance (most often green then) translates into chroma errors.

When a Bayer image is well behaved, i.e has low levels of noise, a good Bayer-interpolation algorithm can extract surprising amounts of real (as in - was actually there in front of the lens) detail. But this is a numerically quite unstable process with low thresholds of hysteresis. This means that when the raw data starts to deteriorate, include more errors - noise - those errors spread out into the surrounding 8 pixels, messing up both their chroma and hence also their luma.

This makes the blotchy noise from older raw converters.

You can see this even more clearly in Canon sensors, that have a very low differentiation between green and red in raw color - older raw converters that didn't handle this well translated errors in that weak region into green-magenta blotch noise. A camera with more balanced channel separation (better? maybe, maybe not...) like the Sony A900, the Kodak DCS or the Leicas have a much more even spread in color noise. Noise came in all colors, not predominantly "just" green/magenta.

Jul 01, 2013 at 09:12 AM

alundeb
Online
• • • • •
Upload & Sell: Off

p.2 #4 · p.2 #4 · My X-E1 vs D7100 High ISO comps

I get what you are saying, but there is nothing (except the higher green coverage) in the X-trans layout that makes it possible to extract luminance with less interpolation.

With Bayer, we can choose if we want to trade color resolution for noise, and normally choose not to.

With X-trans we have no choice but to trade color resolution for noise.

Claiming X-trans is better than Bayer for high ISO is a marketing gimmick.

Jul 01, 2013 at 10:31 AM

aleksanderpolo
Offline
• • •
Upload & Sell: Off

p.2 #5 · p.2 #5 · My X-E1 vs D7100 High ISO comps

If I am not mistaken, I think theSuede is saying is that when you average or sample over a larger area, you will have less color noise especially under low light condition.

That said I agree with alundeb, I would rather have the choice in processing. The color noise removal in LR is pretty effective for my use. I do not know how to get rid of that color bleeding though. It's almost like the red and blue channels are operating at a lower resolution than the green/luma.

P.S. One thing that is interesting to me is why there is such a big difference in underexposing at base ISO and pushing (or pulling?) in LR vs high ISO for D7100. To my untrained mind they are both amplifying signal, one in camera and one in computer?

Jul 01, 2013 at 11:05 AM

snapsy
Online
• • • • • •
Upload & Sell: On

p.2 #6 · p.2 #6 · My X-E1 vs D7100 High ISO comps

justruss wrote:
Snapsy, that last link indicates to me: Better to shoot at 6400 and push in post, than shoot at 25,600 (which is, I assume, pushed in-camera really).

I can then apply whatever amount of smoothing/noise reduction I want afterwards. The "native" 25,600 shot is a blurry, smooshy mess.

Yep, which is pretty much what you have to do anyway since the X-E1 doesn't let you go above ISO 6,400 for raw. It only supports ISO 25,600 for JPEGs.

Jul 01, 2013 at 12:10 PM

snapsy
Online
• • • • • •
Upload & Sell: On

p.2 #7 · p.2 #7 · My X-E1 vs D7100 High ISO comps

Jul 01, 2013 at 12:14 PM

Emacs
Offline
• •
[X]

p.2 #8 · p.2 #8 · My X-E1 vs D7100 High ISO comps

Could you please share RAW files of the top scene and ISO6400?

Jul 01, 2013 at 12:55 PM

zhangyue
Offline
• • • • •
Upload & Sell: On

p.2 #9 · p.2 #9 · My X-E1 vs D7100 High ISO comps

aleksanderpolo wrote:
P.S. One thing that is interesting to me is why there is such a big difference in underexposing at base ISO and pushing (or pulling?) in LR vs high ISO for D7100. To my untrained mind they are both amplifying signal, one in camera and one in computer?

‘I think’ this is a lot complicated than it looks. And really case by case not only depend on camera sensor but also depend on particular picture.

ISO is amplifier in analog domain, Push after capture in raw convertor is in digital domain. High ISO is the gain of amplifier usually maximize the signal, but will degrade Signal to Noise ratio because itself introduce noise as well, maximized signal can utilize more bit in Digital later for process. However, High ISO, not only amplifier signal, but also noise. Once this amplified noise plus noise of amplifier itself more than LSB(smallest step) in digital, you will see noise performance be worse, as you keep feed ADC ‘bad’ signal.

Once the signal through ADC, in digital domain, we no longer improve or degrade SNR.
In most online samples, We focus ourselves in dark area of photo, thus, after certain ISO point, we warrant ourselves see worse performance like Banding, Blotchy….

So, as a guide line, after certain ISO for certain camera, you will not see benefit of push ISO. thus ISO25600, or ISO 128000000 is really market BS with no real meaning other than review your image.

However, in real word, there is possibility that depend on your scene, high key, low key, noise character, or your subject matter, you might still benefit with push ISO than push afterward in RAW. That way you utilized most bits in your digital architecture. With advanced process power of raw converter like noise reduction, tonality mapping, you might get better results.

Jul 01, 2013 at 01:56 PM

alundeb
Online
• • • • •
Upload & Sell: Off

p.2 #10 · p.2 #10 · My X-E1 vs D7100 High ISO comps

It is safe to assume that no cameras have analog amplification beyond ISO 6400. With a 12 bit RAW file, you already use several bits to "quantize" a photon there.

Jul 01, 2013 at 02:10 PM

snapsy
Online
• • • • • •
Upload & Sell: On

p.2 #11 · p.2 #11 · My X-E1 vs D7100 High ISO comps

Emacs wrote:
Could you please share RAW files of the top scene and ISO6400?

Here are the ISO 6400 raws. This link is only good for 5 days.

Jul 01, 2013 at 03:16 PM

aleksanderpolo
Offline
• • •
Upload & Sell: Off

p.2 #12 · p.2 #12 · My X-E1 vs D7100 High ISO comps

Thanks for the link, yours is the definite high ISO test. Surprised that the camera gathered so many "high ISO blow away competition" comment but not many bother to really look to see if it is true.

http://farm8.staticflickr.com/7300/9187799016_300ea770fb_o.jpg

Jul 01, 2013 at 03:48 PM

douglasf13
Offline
• • • • •
Upload & Sell: Off

p.2 #13 · p.2 #13 · My X-E1 vs D7100 High ISO comps

aleksanderpolo wrote:
Thanks for the link, yours is the definite high ISO test. Surprised that the camera gathered so many "high ISO blow away competition" comment but not many bother to really look to see if it is true.

http://farm8.staticflickr.com/7300/9187799016_300ea770fb_o.jpg

I think it is Fuji's understated ISO values that cause the confusion and make the camera appear to be better at high ISO than it really is.

Jul 01, 2013 at 03:56 PM

aleksanderpolo
Offline
• • •
Upload & Sell: Off

p.2 #14 · p.2 #14 · My X-E1 vs D7100 High ISO comps

Ya I think so too. If they can be more honest and make less outrageous marketing claim, then at least I can decide for myself the pros and cons...

And I was going to recommend the D7100 to my friend (I didn't read up about the banding issue, and thought that it is similar to the Sony's 24MP), so this thread is definitely a great help. Thanks OP.

douglasf13 wrote:
I think it is Fuji's understated ISO values that cause the confusion and make the camera appear to be better at high ISO than it really is.

Jul 01, 2013 at 04:02 PM

theSuede
Offline
• • • •
Upload & Sell: Off

p.2 #15 · p.2 #15 · My X-E1 vs D7100 High ISO comps

alundeb wrote:
It is safe to assume that no cameras have analog amplification beyond ISO 6400. With a 12 bit RAW file, you already use several bits to "quantize" a photon there.

Most Exmor "type" sensors end at ~ISO1000 in analog gain. Some off-chip ADC based systems have analog gains up to 3200 (last generation FF) and 1Dx and D4 have ~6400.

Assuming 12-bit files (4096 levels) most -better- cameras in APS get 1photon>1bit at approximately ISO1000, so no damage is done. 14-bit gets you there at ISO250 (though the slower readout arguably makes a difference all the way up to ISO400). At ISO800 the 14-bit file can only be MEASURED to be "better" in the last Ev or so, at which point we're already well down below shot noise. Pattern noise makes a much bigger difference at this point. In some cases 3-5x bigger difference.
...........

Regarding numerical instability; the problem is mainly that we use the SAME interpolation algorithms for both ends of the probability curve, both at low ISO (good reliability) and very high ISO (low data reliability).
Otherwise, if Bayer used the same lowered chroma resolution demand, the difference would be insignificant. If you lower chroma resolution demand by 50% you get a 50% reduction in luma "spill-over" errors. The blue and in some cameras also the red channel are only chroma-carriers with very little luma contrast information, but they still affect luma in a rather significant way if you try to interpolate in a way that maximizes output spatial resolution.

Canon handles this rather brilliantly with the raw noise reduction in the 1Dx, I as a scientist/techie person just wish the function was defeatable... Otherwise I wouldn't mind that type of noise reduction at all. As the signal approaches zero, the chroma influence is lowered, R and B also are allowed much less luma importance in the interpolation, and are preconditioned to "meet" with the average green/average chroma of the area. So the raw image so to say gets "self-stabilizing" again in the raw converter. This makes for much cleaner shadows at high ISOs, the (almost) only thing you lose is chroma spatial resolution - which is insignificant at ISO12800 in the shadows anyway...

Jul 01, 2013 at 04:07 PM

theSuede
Offline
• • • •
Upload & Sell: Off

p.2 #16 · p.2 #16 · My X-E1 vs D7100 High ISO comps

Slight simplification or shortening: the problem NOW is that we apply noise reduction AFTER the interpolation engine has had its' run over the raw image.

This allows the chroma errors to propagate into luma, and what this ADDED to the visible luma noise will remain - even after/if you apply chroma noise reduction in the finished image.

Jul 01, 2013 at 04:11 PM

snapsy
Online
• • • • • •
Upload & Sell: On

p.2 #17 · p.2 #17 · My X-E1 vs D7100 High ISO comps

theSuede wrote:
Slight simplification or shortening: the problem NOW is that we apply noise reduction AFTER the interpolation engine has had its' run over the raw image.

This allows the chroma errors to propagate into luma, and what this ADDED to the visible luma noise will remain - even after/if you apply chroma noise reduction in the finished image.

Thanks theSuede. For your two points regarding NR in this post and before, do you know which part of ACR's pipeline NR is applied, particularly with respect to the interpolation? And that the interpolation is indeed blind to the ISO of the particular file it's rendering?

Jul 01, 2013 at 04:28 PM

snapsy
Online
• • • • • •
Upload & Sell: On

p.2 #18 · p.2 #18 · My X-E1 vs D7100 High ISO comps

aleksanderpolo wrote:
Thanks for the link, yours is the definite high ISO test. Surprised that the camera gathered so many "high ISO blow away competition" comment but not many bother to really look to see if it is true.

Thanks. I was very curious about the reports as well, so much so that I bought the camera only to execute this comparison. I couldn't find any well-executed High ISO comparisons for it online, in particular one that used the same lens and exposure. My curiosity cost me $100 in buying and reselling the camera

Jul 01, 2013 at 10:17 PM

nehemiahphoto
Online
• • • •
Upload & Sell: On

p.2 #19 · p.2 #19 · My X-E1 vs D7100 High ISO comps

After reading this extremely helpful and illuminating thread (thank all who contributed) I am wondering how the X-A1 (non x-trans) compares..similar to typical Bayer?

Are we going to see better color resolution at base ISO and worse high ISO?

Feb 26, 2014 at 08:22 PM

mawz
Offline
• • • • • •
Upload & Sell: Off

p.2 #20 · p.2 #20 · My X-E1 vs D7100 High ISO comps

nehemiahphoto wrote:
After reading this extremely helpful and illuminating thread (thank all who contributed) I am wondering how the X-A1 (non x-trans) compares..similar to typical Bayer?

Are we going to see better color resolution at base ISO and worse high ISO?

The X-A1 uses the exact same bayer 16MP Sony sensor used in all the other 16MP cameras without on-sensor PDAF.

Aside from a distinctly Fuji colour signature, it performs exactly as you'd expect. High ISO is pretty good (looks a bit better than the NEX-5's) but that's probably just a better optimized processing engine, Sony's known to get the worst performance out of their own sensors (Pentax seems to get the best).

Feb 26, 2014 at 09:59 PM

1 2 end
FM Forums \| Leica & Alternative Gear \| Join Upload & Sell 1 2 end

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.