 |
 | Search this Thread |
| 06-23-2009, 05:14 AM | #47 |
 Originally posted by rawr  Hi all. What do you suppose the K7's high ISO performance will be like? I'm curious since the high-ISO of the K20D and K200D (my current digital Pentax) is OK but not great, and this is proving to be a real obstacle to the kind of photography I like to do (street/ night/ nature). I can work around current high ISO limitations with good camera craft, fast lenses, PP etc but quality high ISO performance gives valuable headroom. From the DXOMark.com stats I've compiled a chart (see below) showing the relative performance of these Pentaxes visavis other APS-C sensors and current full-frame DSLR's. So far the Nikon D90 leads the APS-C pack and I must admit I am looking quite closely at it at the moment as a potential future upgrade.. I'd like the K7 to give the D90 a run for it's money in terms of high ISO, because I like everything else about the K7 and Pentax. Does the K7 do high ISO significantly better than for eg the K20, or indeed up to the standard of the D90? Looking at the curves and using the "compare" feature in "print" mode (to somehow take into account resolution), you'll see that the K20's SNR is really equivalent to the one of the touted D300 and really close to the D90's (maybe 0,5 stop lower at most) so really, if we take as a starting point that the K-7 SNR characteristics won't be worse than the K20's, the performances will be as good as the D300 and very close to the D90. If you want to get more than a stop worth of SNR, you have to go FF. But there is one point where the K20D is markedly weaker than D90 (and 40D by the way) and that is DR at mid-ISO (200-1000) and that explains the higher "noise" score of the D90: it's not that the K20 has more noise, it's that it has less DR. | |
| 06-23-2009, 05:56 AM | #48 |
| Originally posted by lol101 But there is one point where the K20D is markedly weaker than D90 (and 40D by the way) and that is DR at mid-ISO (200-1000) and that explains the higher "noise" score of the D90: it's not that the K20 has more noise, it's that it has less DR. I think one of the issues many people have had in reading the low-light ISO scores on the chart, and the current Pentax APS-C sensor placements along the chart in particular  , is that the low-light ISO number you see in the chart is not just about 'simple' noise.If you go and look into how Dxomark craft that score, the ability of the camera to hold onto a certain dynamic range as ISO increases is a key part of the test - ie Low-Light ISO is then the highest ISO setting for the camera such that the SNR reaches this 30dB value while keeping a good dynamic range of 9EVs and a color depth of 18bits. So yeah, the D90/D5000's sensor is able to hang onto that 9EV of DR very well as you boost the ISO, and that characteristic must indeed have helped those two modern Nikon APS-C sensors score as well as they did. | |
| 06-23-2009, 12:59 PM | #49 |
|
The problem is that when they define Dynamic Range, they do that based on SNR (giving what one could call a "double dipping" in their score Their own definition: 3. Dynamic range Dynamic range is defined as the ratio between the highest and lowest gray luminance a sensor can capture. However, the lowest gray luminance makes sense only if it is not drowned by noise, thus this lower boundary is defined as the gray luminance for which the SNR is larger than 1. The dynamic range is a ratio of gray luminance; it has no defined unit per se, but it can be expressed in Ev, or f-stops. Thus, if the camera is utilizing some noise reduction to improve the SNR, it will give a better DR score, even if it lead to some loss of "texture" in the image. Many do like the K20D images better (looking at examples from SpartanWarrior I prefer the K20D). And actually, their way of measuring DR will give as consequence that if you ran an image through e.g. Noise Ninja, the DR score as specified by DxO will improve. I have to side with Dr. Falconeye (yes, he is a physicist, and I have found his posting based on very solid foundations and logic) Originally posted by falconeye: They managed to produce a bag of numbers they don't properly understand themselves (e.g., it took them ages to start to realize that number of pixels does play a rôle ...). Let me say just this much: those DxO physicists wouldn't be able to publish their stuff in a renowed physical publication. Haakan Last edited by Haakan; 06-23-2009 at 01:36 PM. Reason: Corrected a misspelling | |
| 06-23-2009, 01:16 PM | #50 |
| Originally posted by rawr etc etc etc. I don't think you've been reading enough from their site. PS - I don't work for them and have never even used any of their products. But I welcome their contribution to putting useful data out there on DSLR sensors. In this regard I use dxomark in the same way I for example use the photozone.de tests for lens assessments. They attempt to isolate noise with gray scale tests, presumably so they know what "right" is. They measure tonal range and dynamic range while ignoring noise, again, presumably so they know what "right" is. I'm not sure I agree with that. Since much noise is inherently electrical bleed and interference, I'm not sure a flat signal to noise ratio means anything, nor am I sure that the number of colors distinguishable reflects with any kind of accuracy the number of colors that ought to be present. And I'm DEFINITELY not sure their score calculation is anything but gibberish. It's got 90% of one metric, 97% of the other, and 27% of the third, so clearly it's 89% as sensitive! Er, eh? | |
| 06-23-2009, 02:24 PM | #51 |
| Originally posted by Haakan The problem is that when they define Dynamic Range, they do that based on SNR (giving what one could call a "double dipping" in their score Their own definition: 3. Dynamic range Dynamic range is defined as the ratio between the highest and lowest gray luminance a sensor can capture. However, the lowest gray luminance makes sense only if it is not drowned by noise, thus this lower boundary is defined as the gray luminance for which the SNR is larger than 1. The dynamic range is a ratio of gray luminance; it has no defined unit per se, but it can be expressed in Ev, or f-stops. Thus, if the camera is utilizing some noise reduction to improve the SNR, it will give a better DR score, even if it lead to some loss of "texture" in the image. Many do like the K20D images better (looking at examples from SpartanWarrior I prefer the K20D). And actually, their way of measuring DR will give as consequence that if you ran an image through e.g. Noise Ninja, the DR score as specified by DxO will improve. I have to side with Dr. Falconeye (yes, he is a physicist, and I have found his posting based on very solid foundations and logic) Best regards, Haakan The loss of texture can easily be seen on jpegs because other manufacturers have decided to do a rather large amount of NR at their default settings in jpegs while Pentax chose to do none (they actually changed their mind on the K-7 from what I gather) but when comparing raw files in the same converter and with 0 NR and sharpening, detail levels are really similar with a small advantage to the highest MP cameras. Furthermore, their definition of DR looks coherent and it is clear that DR is limited (on the lower side) by noise: the noisier the shadows, the less DR in that region. Noise and DR are related although not equivalent. | |
| 06-23-2009, 02:38 PM | #52 |
| Looking beyond DxO
I have criticized DxO some posts above (not their numbers, but I questioned their ability to properly interpret them). Because it is so much easier to criticize than to make a better proposal, I decided to share some bits of a discussion I had a year ago in a German forum, cf. DigitalFotoNetz.de :: Thema anzeigen - Fomag-Test: Eingangsdynamik und Rauschniveau where I sketched an alternative way of evaluating IQ. For easier reference, I later fetched my contributions into a PDF (cf. attachment). Please be aware that this is all in German language and for non speakers of the language, it really isn't worth to try to make sense out of all this  I rather summarize it for you now.1. State of the art, incl. DxO: noise is considered at the pixel level and measurbated in all possible ways. I.e., noise is only considered at one possible spatial frequency which isn't even typical or standard, but it is the highest possible, the Nyquist frequency. 2. Critics: The assumption that the cross-correlation function for pixel noise is a delta function is naive. On the good part, let's honor DxO for measuring the cross-correlation function for pixel noise. On the bad part, they would just assume that it is noise reduction applied to raw if they find a deviation. They never ever actually measure a signal ... My other critics is that it is ridiculous (from a physicist's point of view) that a serious measure for IQ would depend on the existence of a digital sensor at all. What about IQ for film? 3. My proposition: Have a standard test chart with a fractal pattern where the signal covers all spatial frequencies. Then have the image mathematically transformed into spatial frequency space and use it to normalize a "constant or full signal". Now, take photographs of the test pattern, transform, normalize and compare. Take many such images to determine noise. Don't just use the variation from one pixel to another. Of course, a laser projector (or lens clearly outresolving the spatial frequency integration range) must be used. Then, you get a chart like the following (derived from mathematical considerations):  Blue: Signal curve plotted against spatial frequency (prototypical) Red: Noise curve plotted against spatial frequency (prototypical) Green: Quantization curve plotted against spatial frequency (prototypical) The integrated surface between noise and signal can now be used to define a benchmark for a sensor's (or film's) image quality (at a given ISO step). Please be aware that I totally ignored color here. But on the positive side, my measure would spot banding, clipping the blacks, or other serious and never measured artifacts. [Source: all own science ]When I do criticize DxO, then I mean it at a level they have defined themselves by the amout of effort they claim to put into. And then, I can only say they fail. They fail in a way that you simply cannot seriously say that the sensor with the higher DxOmark is the better sensor. And then, so what? Last edited by falconeye; 06-15-2011 at 05:29 AM. | |
| 06-23-2009, 03:22 PM | #53 |
| 06-23-2009, 03:48 PM | #54 |
| Originally posted by filorp So... simplyfying the case, you would like to stop measure noise as it is - because it wont tell to us about quality of the sensor anyway, and instead you want to measure resolution at the different iso...., and on this ground to determine the real quality of the sensor.... what kind of calculation involve it? I would like to measure the accuracy of reproduction of all (spatial) frequencies, very much like what is done in measuring high end audio. Normally, you have a drop-off of reproduction at very low and very high frequencies. Both are disturbing in images (i.e., banding, unequal exposure, pixel noise and flocking). So, it is not about resolution as such. It is about reproduction of high resolving as well as low resolving patterns. The SNR of low resolving patterns can easily be in excess of 20 Bit! (its all in the German PDF  ) But who did ever measure this?Of course, max. resolution at various iso would automatically be involved because it is the point in spatial frequency where noise and signal curves meet (and therefore, intergration of my IQ mark would have to stop -- so normally, higher resolution at a given iso should mean better IQ in my definition, if noise at lower resolutions isn't too heavy). | |
| 06-23-2009, 09:03 PM | #55 |
| Last edited by rawr; 06-23-2009 at 09:05 PM. Reason: typos | |
| 06-24-2009, 03:25 AM | #56 |
| Originally posted by rawr This conversation is heading in a direction that is exceeding my ability to comprehend it, alas. Originally posted by rawr falconeye's comparison of the issue with 'measuring high-end audio' is a signal to me about where this discussion seems to be heading - into the land of 'audiophiles' seriously discussing dubious matters like the pros and cons of US$10000 2m speaker cables. I am with you on your comment that many audiophiles have lost solid ground since. | |
| 06-24-2009, 09:10 AM | #57 |
|
It is interesting to note from the PhotographyBLOG review of the K7 that the main reasons for the late-breaking changes to the K7 firmware/sensor are high-ISO performance related: Quote: Pentax's engineers have apparently "made a minor change to further improve the performance of the sensor on the camera to be included on the final retail version. As far as we are given to understand this will primarily reduce the amount of noise within the image; especially at higher ISO settings." | |
| 06-24-2009, 11:54 AM | #58 |
| Originally posted by lol101: We are talking raw here and I don't think Canon or Nikon do anymore (or any less) "non-adjustable" NR on their raw files than Pentax. IMHO I would rather have the RAW signal as "untreated" as possible, since I think that it is easier to upgrade external NR software, and getting more powerful processing in my computer, than getting these upgrades into my camera at a later stage, even if it takes its toll on the DxO score. So if the allegations in DPR forum are true that the improved score for D90 is due to doing non-removable NR in RAW (other than the NR that is not possible to do outside the camera), I hope that Pentax does not go the same route for K-7, but rather keep the RAW as "raw" as possible. (to be honest I am a little torn here since I realize that one one hand a higher DxO score might attract more people to buy Pentax, which is good. Then on the other hand, what if the higher DxO score comes at the expense of lower image quality, which is bad?) Best regards, Haakan Last edited by Haakan; 06-24-2009 at 12:03 PM. | |
| 06-24-2009, 12:23 PM | #59 |
|
I was thrilled to find that my old K10D (introduced in 2006) has a better performance (DxO score) than the Mamiya medium format digital back introduced last year.  6000 dollars saved.  | |
| 06-24-2009, 02:07 PM | #60 |
| Originally posted by falconeye: I would like to measure the accuracy of reproduction of all (spatial) frequencies, very much like what is done in measuring high end audio. Normally, you have a drop-off of reproduction at very low and very high frequencies. Both are disturbing in images (i.e., banding, unequal exposure, pixel noise and flocking). I think it makes sense. To me it seems as the way DxO measure the dynamic range is corresponding to measuring the signal on one frequency (homogenous grey over the sensor) and then measure noise at another frequency. In audio terms such a measure would be very easy to trick by just putting in a low pass filter. Of course the signal of the higher frequencies (= resolution), will also be affected, but that is not measured in the DxO measure. Could a variant be to also use fractal pattern with two grey of different darkness instead of black and white. To my understanding in film, the resolution power of pure black and white patterns are different than for "grey-greyer" patterns, and there could perhaps be similar differences in digital. This would give an understanding of the possibility to properly capture more subtle textures. Best regards, Haakan | |
|
| Bookmarks |
| Tags - Make this thread easier to find by adding keywords to it! |
| aps-c, camera, d90, dslr, iso, k7, k7 high iso, pentax, performance, photography |
Similar Threads | ||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| There is more to the K-5 than high ISO, right? | jake14mw | Pentax K-5 & K-5 II | 5 | 10-16-2010 09:33 AM |
| People High ISO K7 | milesy | Post Your Photos! | 6 | 10-11-2010 01:36 AM |
| K7 at high ISO | zelovoc | Pentax DSLR Discussion | 30 | 09-14-2010 07:23 AM |
| K-7 high ISO vs K20D high ISO | supa007 | Pentax DSLR Discussion | 72 | 05-10-2010 04:24 PM |
| Night photography with K10D - High ISO short exposure VS Low ISO long exposure | pw-pix | Pentax DSLR Discussion | 10 | 02-03-2008 01:37 AM |
