the fuji s5pro. it beats even the D3 i think by 2 stops
most likely they bias against fujifilm.. or they know the s5 pro will be high up above the rest..:bsmilie:
Max Dynamic Range - Which DSLR ?
- Thread starter metalgear
- Start date
- Status
most likely they bias against fujifilm.. or they know the s5 pro will be high up above the rest..:bsmilie:
LittleWolf
New Member
AFAIK, Fuji's sensors are peculiar in that they use a mixture of different pixel sizes. That makes it difficult to give an objective and comparable assessment of dynamic range - you somehow have to combine two different sets of performance parameters into one, which almost inevitably requires some arbitrary (i.e. biased) choice on how to do this.
I find the results in the DXO site more reasonable than the clarkvision site.
I think it is reasonable to assume that the D90 sensor is better than the D300 sensor since it is launched later. U just spew out that DXO is garbage but no concrete criticism about their analysis methods.
The DXO site ranked the canon 5D DR to be significantly lower than the D3 while the clarkvision site ranks the 5D higher.
The clarkvision site ranks the D3 SNR to be around 10db higher and yet the D3 DR to be around 1/2 stop lower. The reason for this abnormality is that they calculate DR by individual pixels, ie max capacity divided by single pixel noise floor. DXO calculates it by using SNR instead of noise floor which I think is the correct method.
I think it is reasonable to assume that the D90 sensor is better than the D300 sensor since it is launched later. U just spew out that DXO is garbage but no concrete criticism about their analysis methods.
The DXO site ranked the canon 5D DR to be significantly lower than the D3 while the clarkvision site ranks the 5D higher.
The clarkvision site ranks the D3 SNR to be around 10db higher and yet the D3 DR to be around 1/2 stop lower. The reason for this abnormality is that they calculate DR by individual pixels, ie max capacity divided by single pixel noise floor. DXO calculates it by using SNR instead of noise floor which I think is the correct method.
simranjits
Deregistered
I wonder why does doodah think Dxo think that they never used qualified optical scientists and engineers? The do design some good software that require this kind of expertise.
A later launch date has no guarantee the sensor is significantly improved. For example, the D80 sensor is identical to the one on the D200 (except for faster readout electronics on the D200), as pointed out explicitly by Thom Hogan, a Nikon user and expert. In fact, the D90 sensor is suspected by DPReview to be of lower quality (much stronger AA filter) than the one on the D300 due to the lower cost.
Signal to noise ratio: the comparison of a given signal to the amount of background noise. Background noise is dominated by photon shot noise (especially when the light level is low) and sensor read noise (due to electronics).
Dynamic range: ratio between the largest recordable signal and the smallest detectable signal. While the smallest detectable signal is intimately related to noise, the largest recordable signal cannot be assumed to be equal to the maximum possible raw level and has to be measured.
These two parameters are somewhat related but not the same. For example, the Fuji S3 sensor has tremendous dynamic range at the lowest ISO but puts up a dodgy performance at high ISO (e.g., poor signal to noise due to on-board electronics amplification noise etc).
Analysis by Clarkvision and Emil Martinec sites are based solely on numbers spewed out by the sensor in the RAW files. There is no interpretation of the data by any intermediate software. DXO sites base their analysis on software written BY THEMSELVES to interpret the RAW files. Do you expect CaptureNX and Adobe Camera RAW to output the same images just because they both read RAW files? Also, you do not need to understand sensor electronics and optical physics to write software.
Last edited:
I am not sure why D300 is measured lousier. It could be due to a lemon D300 they have but I do not find fault in their measurement techniques. I agree a later launch date doesn't mean better sensor. But u have not give any reason why the D300 should not be measured worse than the D90. I feel it is entire POSSIBLE and REASONABLE.
U still have not mentioned a SINGLE point on any fault in DXO measurement technique. Just because they write software to read the raw files doesn't mean their software is wrong. I am sure Clarkvision use some software to read the raw files..... u need a software for extraction of data.
u are missing my point regarding the DR measurement. The clarkvision DR measurement is WRONG, IMHO, because they ignore the sensor as a whole. They are measuring the INDIVIDUAL pixels DR by using individual pixel noise floor. I am not concerned about individual pixel DR. I am concerned about the sensor DR. The DXO site takes the SNR of the SENSOR as the noise floor. They set a bottomline SNR which they feel that the noise overwhelms the signal obtained from the sensor as the floor. That I feel is MUCH more applicable to us photographers.
See I have quoted u a point on why I feel the measurement technique of clarkvisions is not "correct" or at least not useful. U have not shown anything wrong about the DXO site.
U still have not mentioned a SINGLE point on any fault in DXO measurement technique. Just because they write software to read the raw files doesn't mean their software is wrong. I am sure Clarkvision use some software to read the raw files..... u need a software for extraction of data.
u are missing my point regarding the DR measurement. The clarkvision DR measurement is WRONG, IMHO, because they ignore the sensor as a whole. They are measuring the INDIVIDUAL pixels DR by using individual pixel noise floor. I am not concerned about individual pixel DR. I am concerned about the sensor DR. The DXO site takes the SNR of the SENSOR as the noise floor. They set a bottomline SNR which they feel that the noise overwhelms the signal obtained from the sensor as the floor. That I feel is MUCH more applicable to us photographers.
See I have quoted u a point on why I feel the measurement technique of clarkvisions is not "correct" or at least not useful. U have not shown anything wrong about the DXO site.
Last edited:
simranjits
Deregistered
So you are saying dxo designed their software with no knowledge of optical physics, right!
got no fuji s5pro leh
when u put $1K plus camera compare to those $4k/5k/6K, who would want to buy those cams? they also need to earn money to keep the website up.:sweat::sweat:
Last edited:
Based on this argument, ANYthing is possible. I can also claim the D60 sensor is many times better than that on the D200.
There is a difference between software used to extract numerical data without interpretation (e.g. DC RAW) and software that extracts the data and then interprets them (e.g. Adobe Camera RAW, DXO etc). That DXO software does its own interpretation of the data (in order to construct an image) is also acknowledged here: "The demosaicing algorithm has to mix information from different color channels, which leads to a combination of noises as well. Sharpening filters may also increase noise, although Image Signal Processors (ISPs) usually include denoising filtering."
Let me reiterate: SNR does NOT tell you ANYTHING about the MAXIMUM possible recorded signal. Get it??? Dynamic range is NOT only about the noise floor but the MAXIMUM possible recorded signal.
Firstly, neither you nor I know the intricate details of how the actual measurements are carried out. Secondly, YOUR definition (not even DXO's) of arbitrarily setting a bottomline noise floor based on what you FEEL is totally unscientific and NOT mentioned anywhere in the DXO site.
simranjits
Deregistered
You yourself are doing your own interpretation of the data, instead of that , the dxo software does it. So maybe you don't trust it and you wish to intrepret the data differently. Any reason why ?
Good question. At first glance, I thought the DXO results seemed pretty interesting. Then, I looked into the details... that's when things get ugly and don't gel with REAL comparison images shown elsewhere. An obvious example is when they claim the D90 is significantly better than the D300 at low light ISO (977 vs 679). At this score, the D90 is pretty close to the 1D3.
Well, if the D90 sensor is so good at high ISO, then there's absolutely NO REASON for Nikon to apply strong noise reduction in their jpeg output and stop at ISO 6400, right? Then, let's take a good look at the jpeg comparison against the D300 here. And this is what it says: The level of detail retained seems comparable with the D300 at ISO 1600 and above (presumably because noise reduction is blurring away any of the super-fine detail that the D300 captures but that eludes the D90).
Elsewhere in the review, you will find this: "the D90 is willing to try its hand at ISO 3200 and even offer a boosted 6400 equivalent mode. These results are less impressive, with the 3200 setting finally sucuumbing to noise reduction blurring and the 6400 equivalent showing a delightful combination of noise, noise reduction smearing and substantially reduced contrast."
There are plenty of examples elsewhere (e.g. Imaging Resource etc etc) that CLEARLY disprove DXO's results.
This is why despite their occasional shortcomings etc, people keep coming back to sites like DPReview and Imaging Resource. Reviewers on these sites interpret their results very carefully. Lab test results are compared against real world images, RAW images are used to check jpeg output, etc etc etc. Pisduck talked about real world photography... well, these sites give REAL world examples that completely contradict the nonsense spewed out by DXO site.
Last edited:
no one replying abt fuji. funny...
ya that is the point. What I am saying is that u do not have any criticism for their measurement techniques but rather keep harping on one Anomaly point. A single error point could be due to a lot of things. The D300 they measured could be a lemon. Or the D90, a terrific sample, etc.
???? Where did u lift that from? u are TOTALLY misreading it.
here is the link where u lift it from.
http://www.dxomark.com/index.php/eng/Technologies/Measurement-definitions/Test-Protocol
What DXO is saying is that they DON'T demosaic it..... duh.... They read the raw data as raw data and do not do RGB conversion as like they say, RGB conversion will pollute the raw data so analysis is done BEFORE That.
who says SNR tells u about the maximum side?? I know SNR is about noise floor and maximum possible signal. Do you have an issue with the way DXOMARK does the high end measurement of DR? I am pretty ok with both clarkvision and DXOMARK way of measuring the high end so I only mentioned the low end side of DR. I feel the clarkvision's way of measuring the low end is wrong.
mentioned.
http://www.dxomark.com/index.php/eng/Technologies/Measurement-definitions/Noise
"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."
different people has different threshold of what is acceptable noise. This is commonly accepted. As long as the standard of noise floor is maintained throughout all comparison, it is ok. There is NO WAY around this.
PS: it is nice to actually READ and UNDERSTAND something before blasting it as totally useless. I am not saying DXOMARK is the holy grail of raw measurement. There are stuff about their measurements which I am not so sure of but what they are doing, ie objectively measuring the raw data, documenting their measurement techniques in DETAIL, is very commendable.
Last edited:
LittleWolf
New Member
If you're referring to aggregate sensor "scores", they are far from a precise science. Describing a sensor takes a large number of parameters; combining different aspects into a single overall ranking is extremely dubious an boils down to "I think this aspect should be weighed so much relative to another", etc.
Don't you just love how much passion forumers have about Dynamic Range?
frankly i dunno why is everyone so anal abt test results than the actually process of shooting? lets not be dull and talk abt stats okay??
I agree their attempt is commendable. My contention with DXO is their INconsistency. And I believe I have provided sufficient evidence to prove beyond a shadow of doubt that the D90 is NOT that much better than the D300 at high ISO. Mind you, the D90 is NOT even better than the cheapo 450D in DPReview's comparison (actually poorer if you account for the strong AA filter in the D90).
This INconsistency is essentially driving many folks like you and me crazy. There are parts of their results which make sense, but others which are totally contradictory to REAL world photos for which I have given links. So, which parts are we to believe and which parts do we reject? :dunno:
I can think of 2 possible (NOT definitive) reasons for this inconsistency: (i) inconsistent test environment, e.g., lights which are not calibrated/checked for consistent brightness (ii) their software is NOT consistent from one camera to another.
BTW, your point about pixel vs image measurement is not entirely unreasonable. But again, DXO is INconsistent here (if we assume they REALLY care about noise at the IMAGE level). For example, independent measurements have shown the D3 has a read noise of about 10 to 15 ADU at ISO 1600 while the 5D2 has a level of about 10 ADU. Now, keep in mind the 5D2 has nearly double the pixel count of the D3. So if we scale the 5D2 accordingly to 12.1 MP (to keep things at the same IMAGE level), the 5D2 ISO 1600 read noise is equivalent to ~ 8 ADU. This again does not gel with DXO data. We can, of course, again wait for DPReview to do their detailed comparison at the pixel level (since DPReview is a such strong champion of low pixel density and totally has no regards for image level comparison) to debunk DXO's results. :bsmilie:
Last edited:
Actually, I agree with you. So, I shan't talk about this anymore. :bsmilie:
remember dpreview uses data AFTER RGB conversion, same for almost all other review sites. besides, from DXOMARK site, the D90 is only 1-2db better than the 450D in terms of SNR, not that significant an advantage.
regarding the inconsitency, as I said, DXOMARK has a lot of writeups and one of them is their test site which is pretty impressive, ie they EVEN CONTROL the temperature and humidity, so I am pretty sure they control the lighting.
As for 5D2 vs D3, that is one comparison which I am very familiar with as I bought a D700 not long ago. I followed the numerous dpreview forum JPEG comparisons. The common conclusion is 5D2 is better at low iso but roughly EQUAL to D700 at high ISO which fits the DXOMARK results perfectly. 5D2 high ISO is also inflated, ie ISO6400 is actually significantly lower than D700 ISO6400. DXOMARK corrects for such ISO errors.
regarding the inconsitency, as I said, DXOMARK has a lot of writeups and one of them is their test site which is pretty impressive, ie they EVEN CONTROL the temperature and humidity, so I am pretty sure they control the lighting.
As for 5D2 vs D3, that is one comparison which I am very familiar with as I bought a D700 not long ago. I followed the numerous dpreview forum JPEG comparisons. The common conclusion is 5D2 is better at low iso but roughly EQUAL to D700 at high ISO which fits the DXOMARK results perfectly. 5D2 high ISO is also inflated, ie ISO6400 is actually significantly lower than D700 ISO6400. DXOMARK corrects for such ISO errors.
- Status
