Nikon Df General Discussion

145791034

Comments

  • tc88tc88 Posts: 537Member
    edited November 2013
    duplication deleted.
    Post edited by tc88 on
  • AdeAde Posts: 1,071Member
    edited November 2013
    @tc88

    You are correct. In fact, DxO publishes normalized figures for this exact reason. The normalized ISO performance of the D800 is virtually identical to the D4 up to ISO 12,800.

    image

    The D800 also has higher normalized color depth and dynamic range compared to the D4.
    Post edited by Ade on
  • BesoBeso Posts: 464Member
    @tc88

    You are correct. In fact, DxO publishes normalized figures for this exact reason. The normalized ISO performance of the D800 is virtually identical to the D4 up to ISO 12,800.

    The D800 also has higher normalized color depth and dynamic range compared to the D4.
    Thank you for providing the correct information Ade.
    Occasionally a decent image ...
  • MsmotoMsmoto Posts: 5,398Moderator
    Nikon introducing the Df reminds me of a piece of poetry by Stephen Crane

    A man saw a ball of gold in the sky;
    He climbed for it,
    And eventually he achieved it --
    It was clay.

    Now this is the strange part:
    When the man went to the earth
    And looked again,
    Lo, there was the ball of gold.
    Now this is the strange part:
    It was a ball of gold.
    Aye, by the heavens, it was a ball of gold.
    Msmoto, mod
  • tc88tc88 Posts: 537Member
    Ade, thanks for the confirmation. I would not be able to find the graph myself on dxomark.

    Msmoto, that's a very philosophical take I like.
  • KLFKLF Posts: 1Member
    A problem with Nikon is that in trying to maximize sales, they missed out on the fact that there are millions of older Nikon film SLR owners who yearned for a AIS coupling in their DSLRs released 10 years ago but Nikon did not bother to release one. Now we have a wide choice with recent mirrorless Sony and Fujifilm, M4/3 releases and so many have committed to too many camera bodies. So another costly option for many with legacy lenses is just not that attractive anymore. The large and heavy DSLR bodies are just too cumbersome especially for travel.
  • TaoTeJaredTaoTeJared Posts: 1,306Member
    TTJ: great graphics! and very good point about printing on-site. One question: as to on-site events printing I would think you could consider a DX sensor such as the one on the D7000 or D5100 if you don't need high ISO. I have found they are adequate to at least ISO 800 and a refurbished D5100 body should be quite cheap.
    I have looked at those and Ideally what I would prefer is to get a D3200 or something like that and have it dedicated to that set-up but DX isn't there yet for these nasty spots - close though. It seems that this realm is living at ISO800. It is workable, especially if it is just prints but file sizes just clog the whole thing up. It has been around 500-1500 shots flying through the system, batch process, quick edit depending on venue, trying to be uploaded to the web, take payment, and both emailed, and printed.
    You are correct. In fact, DxO publishes normalized figures for this exact reason. The normalized ISO performance of the D800 is virtually identical to the D4 up to ISO 12,800
    Sorry but "Normalized" is a cool-aid term and is anything but that and is pure BS. DxO rapidly released that when Canon was screaming at DxO when the D3/D700, D90, and every other Nikon trounced their models. It is a hypothetical equalizer that reduces images to 8mp image. Basically it turns the readings into JPEG basic files and compares those. And what happens then? Every sensor receives basically the same score within 10% of the base, so none look much better than another, and Dx0 gets manufactures off their back. This is how the D800 magically becomes equal to low light to the D4. Sorry, love my D800, but it is not close to the D4 in low light. When sensors jumped over 20-22mp mark, DxO's test went really wonky posting inflated scores.

    Basically all the arguments people want to use are based on the output of the sensor. Why in the world would you take the output, play and mold it into something it is not. I re-size D800 files every day, It doesn't reduce noise, it smashes it together and makes a bigger blob that may or may not be closer to the color it was suppose to read. Noise around eyebrows (i.e. loss of detail) doesn't magically become better when you re-size, it just becomes a less techno-color blob. And when you crop? Oh yeah DxO "print" doesn't account for that at all - that can work both ways. The "Screen" results give a much better and real world result. The "Normalized" is there to keep companies off their backs, and consumers happy by making them think their sensor isn't much different from another one.

    If you believe it is correct to take every sensor, and re-size them to 8mp and effectively remove 8-28mp of output of data? Then all you really need is to help Nikon with it's compact digi cam sales and skip the DSLRs.
    D800, D300, D50(ir converted), FujiX100, Canon G11, Olympus TG2. Nikon lenses - 24mm 2.8, 35mm 1.8, (5 in all)50mm, 60mm, 85mm 1.8, 105vr, 105 f2.5, 180mm 2.8, 70-200vr1, 24-120vr f4. Tokina 12-24mm, 16-28mm, 28-70mm (angenieux design), 300mm f2.8. Sigma 15mm fisheye. Voigtlander R2 (olive) & R2a, Voigt 35mm 2.5, Zeiss 50mm f/2, Leica 90mm f/4. I know I missed something...
  • donaldejosedonaldejose Posts: 3,866Member
    TTJ: The D400, when it arrives, should be good for about one stop higher ISO than current DX sensors. But I also expect it will be 24mp. A D700 should work fine; if you can find one with a history you can trust. Any more manufacturer reconditioned D700 or D3 models appearing or one someone had and didn't use much?
  • TaoTeJaredTaoTeJared Posts: 1,306Member
    TTJ: The D400, when it arrives, should be good for about one stop higher ISO than current DX sensors. But I also expect it will be 24mp. A D700 should work fine; if you can find one with a history you can trust. Any more manufacturer reconditioned D700 or D3 models appearing or one someone had and didn't use much?
    I have been looking at D3 used bodies since they are rated (and built) higher. Still $2k for them & $1000 more get's me a system that is an upgrade and more uses. I am watching what D700's price drop to, With the D600 fire sale at $1,300, and if the D700 dips below $1,000, then it becomes interesting especially when I know 12mp sensor file sizes race through the system and I can go through 3 of those bodies (as well as utilize all my D300 accessories.) It's all about what is throwing good money at bad, better, good, or great. It is also about all the trade-offs. Oh and also what else I need/want to buy ;)
    The fact remains, when you downsample an image, the noise disappears. Comparing the D800 at 36mp to the D4 at 16mp isn't as fair as seeing what the 36 downsampled to 16 looks like; else, why don't we see how good the 16mp files looks upsampled to 36?
    The fact remains that this camera will be shot at 1600 - 12,800, and the D800's blobs downsized still are blobs.
    D800, D300, D50(ir converted), FujiX100, Canon G11, Olympus TG2. Nikon lenses - 24mm 2.8, 35mm 1.8, (5 in all)50mm, 60mm, 85mm 1.8, 105vr, 105 f2.5, 180mm 2.8, 70-200vr1, 24-120vr f4. Tokina 12-24mm, 16-28mm, 28-70mm (angenieux design), 300mm f2.8. Sigma 15mm fisheye. Voigtlander R2 (olive) & R2a, Voigt 35mm 2.5, Zeiss 50mm f/2, Leica 90mm f/4. I know I missed something...
  • snakebunksnakebunk Posts: 993Member
    @TaoTeJared: The images cannot be the same when they are downsized, yet the noice remains and the D4 is superior. I think you have to make up your mind :)
  • AdeAde Posts: 1,071Member

    If you believe it is correct to take every sensor, and re-size them to 8mp and effectively remove 8-28mp of output of data? Then all you really need is to help Nikon with it's compact digi cam sales and skip the DSLRs.
    Normalization does not remove image data. Normalization combines image data. One cannot make a coherent argument without understanding this difference.
  • WestEndBoyWestEndBoy Posts: 1,456Member
    Ade said:
    Normalization does not remove image data. Normalization combines image data. One cannot make a coherent argument without understanding this difference.

    There is no difference Ade. Combining data removes data. Consider the following:
    1.
    Take two data points - 25 and 50.
    2.
    Combine them by addition..
    3.
    You have 75.
    4.
    You have lost the two data points, because you don't know how you arrived at 75. Eg. You might have combined 5 by 15 by multiplication.

    TaoTejared, you are a bright guy. I have been paying particular attention to your posts. What is your education background?
  • tc88tc88 Posts: 537Member
    edited November 2013
    jshickele, your understanding is quite wrong. I think Ade is going to vomit reading it. :)

    Note the word Ade used is combine, not add.
    Post edited by tc88 on
  • WestEndBoyWestEndBoy Posts: 1,456Member
    Hi TC88, pleasure to meet you.

    I am not quite sure of your meaning. Am I misguided about math or TaoTejared?

    Ade, I am concerned about TC88's comment. Please don't read my post unless you are in the presence of a doctor. Otherwise, I accept no liability.
  • tc88tc88 Posts: 537Member
    jshickele, sorry I was just joking here. :)

    Yes, your math is wrong. If your logic is correct, then combining enough pixels of mid tone values will give a white out picture.

    I will explain later why down sampling improves SNR if no one else has done by then. Got to go for now. :)
  • WestEndBoyWestEndBoy Posts: 1,456Member
    TC88 said:
    I will explain later why down sampling improves SNR if no one else has done by then.

    Thanks, this technical area is quite fascinating and I look forward to it.
  • mikepmikep Posts: 280Member
    it was 5 years ago that canon brought out the 5dmk2, which took a lot of market share based on its video ability. FIVE years !

    and now, in 2013 we have nikon releasing this DF with no video .... while all its cameras for the last 4 years have had video .... under a banner of "pure photography"

    while they have just released a d610, which is better in every way for all practical purposes for 2/3s the price of the DF

    the DF has controls on top of the camera which will require two fingers to change as opposed to any other dslr which will require one finger to rotate the jog-dials (no big deal but reality)

    worst part of the d600 (excluding the oil) was the AF system, which has unfortunately been carried over to the d610, which has been stuffed in this DF also ..... i mean its absolutely ridiculous to put this DX AF system in these FX cameras ....

    and then the price!! haha omg you can get a d800 cheaper than this thing



  • AdeAde Posts: 1,071Member
    @jshickele

    It's really not hard to understand. I will literally try to explain this with elementary school math.

    Suppose we want to take a picture of a uniform wall, with a tonal value of "10" (for super easy math).

    First, we try a 1-pixel camera, which (due to noise) registered 11. The pixel noise error is 1 out of 10, or 10%.

    Second, we try a quadruple resolution, 4-pixel camera. Due to noise, the camera registered [7, 12, 8, 9] for its four pixels. Compared to the actual value of 10, our 4-pixel camera made absolute-value errors of [3, 2, 2, 1] for each respective pixel. So, the average pixel noise error is (3 + 2 + 2 +1) / 4 = 2 out of 10, or 20%.

    At this point, we might conclude that our 4-pixel camera is "noisier" because its average pixel noise error is 20%, compared to 10% of the 1-pixel camera.

    But, we realize, if noise is truly a random process, then over time it tends to average itself out. That is, some noise will produce results which are too low (like [7, 8, 9]) while other noise will produce results which are too high (like [12]).

    So let's try combining our 4-pixels into 1-pixel by using the averaging function: (7+12+8+9)/4 = 9. The error after averaging is now is 1 out of 10, or 10%.

    Therefore, even though our 4-pixel camera had more per-pixel noise (20%), it actually has the same 1-pixel performance as our 1-pixel camera (10%).

    I hope you can see that combining data (via the averaging function in this case) is very different from removing data. If we had simply removed data from our 4-pixel array [7, 12, 8, 9] to arrive at a 1-pixel solution, then on average our error would have remained at 20%.

    With slightly more difficult math (involving logarithms), we can mathematically show that this result can be extended to the normalization of arbitrary megapixels. So whether we are talking about 1-pixel cameras, or 8mp, or 24mp, or 100mp, the same concepts apply.
  • WestEndBoyWestEndBoy Posts: 1,456Member
    But you have removed the data in the original 4 pixels by combining them into one with the average function. Sure you have kept the noise down, but the nose and eyes are now missing and we are left with what TTJ described as a blob (I exaggerate to illustrate the point). I don't have my D800 set to combine pixels (well, I shoot Raw plus Fine JPEG, so unless you count that, I am not aware of a setting that does this).

    The combining (done by averaging) seems to be removing data by reducing the resolution.

    Am I missing something?
  • AdeAde Posts: 1,071Member
    Realize that both hypothetical cameras ended up at a baseline 1-pixel resolution and with the same noise level of 10%. One does not have more details ("nose and eyes") than the other. It's a fallacy to think otherwise.

    Instead of 1-pixel, we can arbitrarily pick any other pixel count baseline (more below).

    You (perhaps unknowingly) combine pixels every time you view an image at less than 100% crop; every time you email or post an image at other than original size; and every time you print. I.e., your image processing software automatically performs this combining for you, using various algorithms (which may or may not be more better than simple averaging, from a noise perspective).

    For example, if you print an image on a 8"x12" paper (20cm by 30cm), with a printer resolution of 300 dpi, then you are really printing an 8.64 MP image (8 * 300 * 12 * 300). Since you have a D800, all that 36 MP goodness will be combined to produce a less-noisy 8.64 MP print, behind the scenes.

    This is exactly why, by the way, DxO picked a normalization baseline of 8.64 MP. But they could have as easily picked another baseline without changing the basic results.

    We've strayed quite a bit from discussing the Nikon Df so I think I'll leave it here.
  • MsmotoMsmoto Posts: 5,398Moderator
    edited November 2013
    If this thread is still about the Df body, another way to think about this is the song "From a Distance".

    And, PitchBlack, yes, the bottom line is we all need a long tele…... :))
    Post edited by Msmoto on
    Msmoto, mod
  • WestEndBoyWestEndBoy Posts: 1,456Member
    Yes, no more to be said. Ade, thanks for your efforts. Your last explanation was something that I understood but I now understand your thinking, reasoning and perspective, so thanks.
  • heartyfisherheartyfisher Posts: 3,192Member
    edited November 2013
    You guys giving up too easily ! :-) let me have a go.. ;-)

    The reason why in practice that theoretical model fails is because the 4 Pixel's total surface area of sensitivity is less than that of the 1 pixel because you lose light at the borders of the pixel ie the inter pixel spaces do not have light sensitivity. Thus the s/n ratio is less even when you do "sum" it back into 1 pixel. so the larger pixels always win if the resolution is sufficient. (assuming all else being equal)
    Post edited by heartyfisher on
    Moments of Light - D610 D7K S5pro 70-200f4 18-200 150f2.8 12-24 18-70 35-70f2.8 : C&C very welcome!
    Being a photographer is a lot like being a Christian: Some people look at you funny but do not see the amazing beauty all around them - heartyfisher.

  • tc88tc88 Posts: 537Member
    edited November 2013
    heartyfisher, the border is really a very small area. So in theory, what you said may be true, but in practice, the effect may not be significant. :)

    But the proof is in the normalized data. If one sensor indeed has better efficiency, let's say a sensor made five years in the future, it should show up as an improvement in the normalized graph regardless the actual native MP.
    Post edited by tc88 on
  • blandbland Posts: 812Member
    @PitchBlack .... if people are buying this camera as a "hey look at me" as you made reference to, they'll obviously be in for an awakening, won't they? I will enjoy the retro look but people I'm shooting for are going to be wondering why in heck did I bring an old outdated camera for. I get it, you don't like it but please spare me and those of us who are getting one from being in the same category as a Harley rider, we ain't the Village People!
Sign In or Register to comment.