Tariq Gibran wrote:
These are standard deviation numbers, not stops (which means little to me since I forgot what these actually mean via statistics!) . According to Pop Photo:
"Using DxO Analyzer software, we measure the standard deviation of grayscale patches across a full range from dark to light."
I have also learned that they do use RAW files and convert them into tiffs using the manufacturers raw converter before analyzing. So, it's nice to know they are at lest not using out of camera jpegs.
Tariq,
Yes, these units are not stops, just some unit of noise measure.
What the guy in the pop-photo article was saying was that he thought the A850 was one stop ahead in high iso noise performance because the noise value of the A850 at iso 400 was the same value as the A900 at 200. Even though that is true at ISO400, you can see by the numbers that at higher iso the improvement is a slight improvement, not a whole stop.
What is not right in the pop-photo tests is the color accuracy tests which do not jive with the idea that the A900 has high hue resolution.
As far as these DXOMark color response numbers of various cameras, you can see that the Phase 65+ and 45+ have numbers in the around 74 and I have never heard anyone complain about their color response. Instead people usually praise MFDB colors. Also, I have never seen much large differences between the various models of Canon cameras even though their color response numbers vary. Look at the color response of the A700 and it is in the 70's but I bet people can not detect very large color differences between it and the A900 because they share the same maker color fingerprint.
Thrice posted a picture from a Sony P&S the other day on the alternative thread and I noticed right away the more vibrant/vivid Sony type colors in the shot. I think the maker's color fingerprint is the more important factor in this perceived better color thing.
I think we all need to be a little careful on how we read and interpret DxO Mark scores...at least guys like me. Obviously experts, like theSuede, have a solid grip on these color measurements and how they relate to real world use. Interestingly, Iliah Borg mentioned something about this a while back.
I would think that different raw converter profiling or jpeg settings could give your 1Dsiii landscapes a very similar "look" to the A900, Wayne. I've always contended, due to what Iliah has said, that the color difference was in blue resolution, and affected skin tones, rather than foliage. I don't see why the 1Dsiii shouldn't be capable of similar greens to the A900, with the right profiling.
wayne seltzer wrote:
What the guy in the pop-photo article was saying was that he thought the A850 was one stop ahead in high iso noise performance because the noise value of the A850 at iso 400 was the same value as the A900 at 200. Even though that is true at ISO400, you can see by the numbers that at higher iso the improvement is a slight improvement, not a whole stop.
What is not right in the pop-photo tests is the color accuracy tests which do not jive with the idea that the A900 has high hue resolution.
Here is exactly what is said in the Pop Photo article:
"From ISO 100 through 400, it showed about a 1-stop improvement—that is, the A850’s noise level at ISO 400 (Very Low) was about the same as the A900’s noise level at ISO 200."
I don't understand the numbers and it may be that they are not directly comparable to other cameras in any sort of absolute way. That is, the numbers only represent deviations of noise using only that camera's characteristics as a base. Reading the above quote tells me that the a850 at ISO 400 equals the a900 at base ISO of 200. That seems like a pretty big deal and the reason for my initial question.
wayne seltzer wrote:
Thrice posted a picture from a Sony P&S the other day on the alternative thread and I noticed right away the more vibrant/vivid Sony type colors in the shot. I think the maker's color fingerprint is the more important factor in this perceived better color thing.
Sony P&S cameras are known for over the top, saturated color. That is certainly not the case with the color out of the a900. P&S camera's in general are geared for consumers that react to over the top, saturated color...just like the default display colors of all those LCD big screen tv's when viewed at BestBuy. In general, as saturation and vibrance go up, distinct subtle nuances of hue go down. This is a fairly easy thing to see manipulating an image in Photoshop.
Tariq Gibran wrote:
Here is exactly what is said in the Pop Photo article:
"From ISO 100 through 400, it showed about a 1-stop improvement—that is, the A850’s noise level at ISO 400 (Very Low) was about the same as the A900’s noise level at ISO 200."
I don't understand the numbers and it may be that they are not directly comparable to other cameras in any sort of absolute way. That is, the numbers only represent deviations of noise using only that camera's characteristics as a base. Reading the above quote tells me that the a850 at ISO 400 equals the a900 at base ISO of 200. That seems like a pretty big deal and the reason for my initial question. ...Show more →
I am pretty sure these numbers are some kind of unit of noise measurement in their test which is comparable across cameras, otherwise their is no use for the test. When they compare cameras they refer to these noise numbers between cameras.
The noise improvement of iso 400 becoming same as iso 200 of A900 may be significant to you but to me and alot of others who feel that the noise levels of these cameras at iso 100-400 is very small and very close to the same level. When someone says a camera has an extra stop better of noise performance, I take that to mean that high ISO's 800 and above continue to be one stop ahead in noise levels.
Of course another thing about these noise tests are that they usually don't resize the image of a smaller MP camera to the size of a bigger MP camera which is not fair. Not sure that pop-photo does this resizing, probably not. This is Richard's fight.
Another thing that I think should be done in these noise tests is to plot the resolution at the same time so you can see if there is some heavy NR going on in the camera before the raw file is generated at high ISOs.
wayne seltzer wrote:
The noise improvement of iso 400 becoming same as iso 200 of A900 may be significant to you but to me and alot of others who feel that the noise levels of these cameras at iso 100-400 is very small and very close to the same level. When someone says a camera has an extra stop better of noise performance, I take that to mean that high ISO's 800 and above continue to be one stop ahead in noise levels.
I don't think you can lump the a900 into a group - "these cameras" - because this particular camera does exhibit more noise than most others you might be thinking of. At ISO 400 and shooting raw, even a little "fill light" type shadow recovery on a raw file will show significant noise. Similarly, if you try to rescue an underexposed ISO 400 shot, you will also run into significant noise - more so than other, full frame cameras. That is why this issue is important to me and the reason for my original post/question.
Ok, but I don't think these generic ISO noise tests give an indication of how much noise there is in the shadows for when you want to rescue a shot underexposed by a stop. This is the very thing that Lloyd Chambers raves about the D3x and how it is another league compared to the 1ds3. Well, I rarely try to rescue pictures that are that underexposed to the point that I see significant noise showing up in the shadows. Chimping and bracketing help avoid that. If you want lighter better shadows, you can use the shadows from a lighter bracketed shot if the subject is static too and for those more HDR type scenes.
Another thing which will help in the future is the color aware metering which is in the 7D and 1d4 right now.
The metering is way more precise now and the WB is much better. No more remembering to dial in extra compensation due to the subject color.I am sure all camera's will have this soon if they don't have it now.
Anyway, if there is a test which gives you an accurate indication of how much noise there is in the shadows for these underexposed cases let me know.Some of that may be helped/worsened by the RAW developer chosen.
wayne seltzer wrote:
Well, I rarely try to rescue pictures that are that underexposed to the point that I see significant noise showing up in the shadows.
The point I'm making is that with another camera, say a 5D, if you do underexpose a shot or require a certain amount of fill light, no problem at ISO 400. Shooting the same shot with the a900 will result in noise where it never did with the 5D. With the a900, accurate exposure is more critical than with other cameras. For me, the other benefits the a900 gives me far outweighs this slight drawback, and I'm not in the habit of underexposing shots by one stop!, but the issue is there. So, once you shoot with an a900, you will see that noise shows up in those shadows much quicker than with what you might be used to seeing. The file is not as "elastic" or malleable at the shadow end but, in exchange, you get a much better highlight response before hitting the point of clipping. Thus, in fllm parlance, with the canon I would always expose for the highlights and NEVER expose to the far right and with the Sony a900, you are better off to expose for the shadows and expose to the right. Almost the difference between shooting chrome or negative film. Again, with this camera, a one stop advantage, even at ISO 400, would be a big deal. If you live with the camera for a while, you can appreciate this.
I understand that you want more margin so that if you have add fill or add exposure that noise doesn't crop up in the shadows. Since I haven't shot the A900, I have no idea how easily this noise crops up compared to the canon. I guess I always shoot to the right without clipping highlights (unless it is a shot with no highlights and I am concerned about the colors getting too saturated/bright, that they don't look good when dialing back down the exposure in LR) and if I feel that I that later I will want to lighten the dark parts of the shot quite a bit then I would bracket a stop lighter and then combine shots in PS.
Anyway, for the case when you are shooting at ISO 400 and you want to boost the exposure 1 stop in the raw converter, then I think you wanted to say that the noise level would be the same as ISO 800. But this is only true for like ISO 50 on canon where the camera is just doing a straight push of one stop. But for this ISO 400 case I think it will be worse than ISO 800, but anyone feel free to correct me.
I am also trying to say that I think you can not ascertain how each camera behaves as far as how much noise is going to show up when you boost the exposure in the shadows by looking at this iso noise test results.
wayne seltzer wrote:
I guess I always shoot to the right without clipping highlights (unless it is a shot with no highlights and I am concerned about the colors getting too saturated/bright, that they don't look good when dialing back down the exposure in LR) and if I feel that I that later I will want to lighten the dark parts of the shot quite a bit then I would bracket a stop lighter and then combine shots in PS.
At least with my 5D, though the lcd histogram might not show highlights are being clipped, they often were. To see this, simply shoot some clouds at various exposures, look at the histogram for each, and then use the info tool in Photoshop to look at the values in the clouds. I bet if you expose to the right, you would find less detail in those clouds and there actually is some clipping going on. That rarely happens with the a900 which has a much more gradual shoulder in the highlights compared to my 5D.
Bracketing and combining shots is a luxury not always possible with moving subjects.
In any case, if there is a noise advantage of one stop for the a850 vs a900, I'm interested in hearing about it!
well you know it isnt like Sony started from scratch in the DSLR world now did it?
they had alot of help along the way,you know BUYING Minolta remember? oh yeah and being in bed with Nikon helps a lot too...
Most of the difference between 900>850 at lower ISOs lies in how they use the on-board amplification, and the power supply lines to the cirquits on board the sensor chip. The 850 has lower amplification at low (base) ISO, and slightly higher at the highest ISO's. What this means in an actual photo is that Sony "fill the well higher" at base ISO in the 850. This gives some compression in the absolute top of the exposure range, as they "push" the well a bit past the linear region.
By "compression" I mean that 2x light amount should ideally result in 2x higher raw data numbers (plus the base line). Raw is linear format, "gamma 1.0". The 850 compresses this a bit as the capacitance that stores photocharge runs out of "room" - to double the raw-ADU in the last available stop before clipping takes MORE than a doubling of the light amount.
What seems most likely to me, and the first thing I will check is how Sony handled the blue/red amplification. This is one of the Achilles heels of the Bayer systems - that you have to give very big margins for WB (you keep the red and blue very weak so that you can "tilt" the spectrum to compensate for WB). A small limitation on the extreme ends gives you better noise. at low ISOs. I know some of the older single-digit Nikons actually had a WB-dependent amplification already in the readout of the sensor wells, but that this system was abandoned because of the very much higher "worst case" error this could induce if you set WB way off (clipping of channels).
I think it was Ilah that first made me test the "magenta filter" trick with digital cameras - and it works REALLY well. It goes something like this: At "normal" WB ranges from 4500-6500K, red and blue are underexposed in the camera by almost one full Ev. If you're at ISO100 in the camera, that's only the green channel. The blue/red are at ISO100 pushed to ISO200 in post to WB the scene.
If you're shooting scenery, or working in a studio with pro-grade strobes (at ISO100) you can trick the system by adding a magenta 020 or even 030 or FLD filter to the lens. By doing this you expose all three channels at the rated ISO, and gain some noise performance. You also get a very marked increase in colour accuracy in the shadows. The trick is finding magenta or FLD filters of sufficient quality, so that you don't loose more than you gain... :-) I used this system for the greater part of the summer, and I see a clear difference. The limitation is that you should NOT be shutter-time limited if you wan't to use this "trick", as it increases shutter-times by 2/3Ev.
Sou you CAN get a noise-lowering effect of almost 2/3 Ev in the A900 in the shadows if you: Are not shutter-time limited; Are not shooting outside the 4500-6500K range of WB.
BTW - the original 5D is an awesome camera. It has it's limitation, but within these limitations it really shines. It also has a lot better ability to give you great colour at lower ISOs than the 5D2.
If you don't mind - what raw-converter are you using right now, and with what profiles? I'm currently in the business of adapting accurate colour to "user preferred" colour in conjuction with some unspecified software brands right now - so I wouldn't mind using you as a lab-rat by sending you some profiles for evaluation. This goes for everyone that I can find the time for here btw - your task in return for the free profiles would be to give a short evaluation of "could be different/better/worse".
Tariq, yes sometimes some slight clipping can happen, and if I am those situations where I am worried about clipping the highlights I will dial in -1/3 compensation and use some recovery in LR if things are still clipped. But yes, I agree you have to be carefull and its interesting to know that the Sony's highlights rolloff more like film and thus more forgiving.
theSuede, thanks again for explaining all this and it makes sense.
I am using LR 2.6 and 3 beta right now as I am PC based but I have been evaluating C1 pro and DPP again after getting the 7D. I have 5D, 1d2n, 1ds3 and now 7D.
I tried some profiles somebody provided on a forum one time but they didn't seem to work that well. I am up for evaluating your color profiles and am excited in the quest for better/accurate color.
I have a question about whether the WB setting set in the camera or determined by the AWB of the camera has any affect on the raw file produced by the camera? No, right?
I am noticing better color right away from my 7D over my 1ds3 due to the new color aware metering system and much better AWB unit. The 7D seems to find something white or neutral in the scene and then determine the correct WB especially in indoor lighting scenes where my 1ds3 is always way off in the AWB it determines so colors look crappy till the correct WB is set in LR.
How do you compare the 1ds3's color to the 5D and 5D2?
And when you say the low iso color is better for the 5d over the 5d2, you are really saying that the metamerism areas are smaller and thus the color is more accurate in those areas but that outside those areas the same Canon color fingerprint makes the colors of both cameras very similar. Correct?
Finally, do you believe my earlier statement that maybe what people perceive as the better color camera is due more to the fingerprint than the camera's hue resolution or better color accuracy in the metamerism areas? The P65+ and P45+ have worse metamerism but people seem to think they have excellent color, right?
Better colour is mostly fingerprint related, yes. We generally look at the overall balances, not the detail. Actually noone except the photographer know the real scene, so "accurate" colour is very misleading unless you're in a position to say what reality was.
The 7D is good overall, but seems to have the same slightly random approach to metamerisms in skin tones (yellow-orange). About the WB I really can't say, since I almost never look at the AWB result. I just trust it to make my histograms in-camera somewhere at least close to real values, so that I can correct exposure if needed.
I think the 1Ds3 is a bit better than the 5D2, there are some small and very subtle differences in approach.
Regarding the PO cameras, the subject of "colour" gets even worse... There's always another layer of information.
What is generally reported as "metamerism index" is not an absolute value. It's a value that compares the metamerisms in sample representation "X" to the inherent metamerisms in the human vision system. We do have a lot of "kinks" in our native perception of colour... The eye isn't "logically linear" - we experience some combinations of wavelengths as something entirely different than what a pure spectral graph would imply. A camera that has exactly the SAME metamerism combinations as the eye is said to fullfill the Luther-Ives condition, and the camera will then render the scene just as a human would "experience" it. It would get a score of "100" in the DxO measurement system, and the metamerism locus (area of error) would be a singular mathematical point at the white-point in the middle of the R-G-B triangle.
But that a system has a lower score can actually also mean that the camera has a HIGHER colour accuracy than the eye, that it resolves things that the eye can't. In the comparative sense, that's also an "error". The A900 also has some minus-points for resolving more than the eye... Having a lower absolute metamerism index than the eye can give "supercharged" colour, and resolve hue differences (light wavelength combinations) where we as humans couldn't distinguish any difference. Is this good or bad? I think it makes correct balance harder sometimes, but that it is a lot better than the other way around.
Coincidentally, I just ran into your "blackbody light" a900 color profile for Lightroom over on another forum, and it looks very good. I'm gonna try it out for a while. Thanks, the Suede.
douglasf13 wrote:
Coincidentally, I just ran into your "blackbody light" a900 color profile for Lightroom over on another forum, and it looks very good. I'm gonna try it out for a while. Thanks, the Suede.
can you post a link? Would love to try this as well.
