Calphate wrote:
Just curious, how did you conduct these tests? Select an area that you considered "black" and check the standard deviation?
His software
Camperjim wrote:
Where did you come up with this concern? Did you read any of the preliminary reviews? It is clear the 90D has a strong AF system.
I don't see a concern about DR to be overblown. It is an important factor for most of us. Fortunately it seems to be substantially improved.
That’s not clear at all. There are indications the AF system in the 90D will be better than the 80D, but until we see more tests, especially with BIF, the performance is unknown. My biggest complaint is the poor tracking ability of my 80D, that has to significantly improve for me to consider upgrading.
Yes, from ISO 1600 they are pretty much equal (and despite not measured, in overall image quality, since Canon sensors are no slouch in light gathering efficiency). Canon usually has been great at high ISO, so it is back on his tracks.
If this is correct, and has been implemented in the 90D ... it looks like it could be showing benefit in the north end. Picking up roughly a full stop, while raising pixels by 33% ain't too shabby.
Imo, you just need to get the ISO in the ballpark to start. You can't seem to play with Canon files the same way as ISO-less approach of gross underexposure and a ton of lift. Get 'em in the ballpark (i.e. less than 2-3 EV lift needed) with ISO, and they tend to do better, than dragging them up from a lower ISO and cranking the lift (4-6 EV).
As such, seeing the DR up a notch in the higher ISO's is encouraging.
This could play well with Auto-ISO approach scenarios.
Given that the 80D has about 1 EV advantage over the 7D2 at base, and is even in the higher ISO's. This would put the 90D approximately 1.25 - 1.00 EV improved over the 7D2 across the entire ISO range. Matching FPS @ 10 (or 11) with the 7D2, the main thing that remains to be seen compared to this being a hybrid of the 7D2 & 80D is going to be AF performance, imo (build & dual card slot, notwithstanding)....Show more →
As somebody else has noticed, the 80D already had in-sensor ADC. In fact, it was the first Canon camera with this feature (maybe that was the reason for its lackluster high ISO DR).
Note that manufacturers are conservative and raising the resolution usually doesn't negatively affects the sensor performance (neither the numbers we report here, because are normalized to a 8-megapixel equivalent sensor) so there is no much merit related with the additional resolution. In fact, read-noise prone sensors benefit in DR from the extra resolution (that is the only reason the 5DS stands out a bit in DR compared with the rest on its crowd).
At low ISO, the 90D has almost 2 full stops of additional dynamic range compared to the 7D2 (1.9 to be exact). At ISO 6400 only around 0.4 EV, but I think the 7D2 was already a great improvement over his predecessor (beware that data in the linked old post is not normalized... the 7D2 numbers published require adding +0.68 EV as the post indicated).
cpe1991 wrote:
A nearly one stop increase in DR under conditions where photon noise is the limiting factor with efficient modern sensors is truly remarkable. That means the sensor has the same DR as a 5DIV with its 2.56x larger sensor, and well above any Sony or Nikon APS-C. Have I got something wrong?
Canon has refined the read noise performance (not yet to the best of the industry) which is what matters for DR and shadows performance. Photon noise sets an unavoidable upper limit in the SNR (signal-to-noise ratio) under good lighting and modern sensors are all very similar there because have similar efficiency.
I haven't look since a lot of time to APS-C sensors (the 90D is a exception due to my Canon roots) but I think that Canon has not yet catched Nikon or Sony APS-C flagships DR (IIRC some of these went beyond 14 EV a long time ago). But at least now has decent figures.
Canon claims the 90D "provides high sensitivity for high ISO capture with reduced noise". Seems like faint praise if there is such a big improvement in the sensor.
Interesting! (I dind't know somebody took the effort to do a video).
To properly calculate the dynamic range, the image must have at least some pixels overexposed. The youtube video uses a image from the 5D (I hope) which was 12 bit, so in that case it is still correctly exposed for the task (upper values close to 4095, but maybe not just at that level, because old sensors didn't get to the very top numbers at ISO 100).
There is a usage example here (which somebody copied from my original post I can't find):
In canon it is very easy to calculate the dynamic range, because the sensor has an area with masked pixels (not receiving light). We can measure the noise on this area on any photo and compare it with the upper (brightest) level in the rest of the image, taking into account that the "0" is usually shifted. The maths used by the application are simple and well explained on the Internet:
range of values 0-16383 (14 bit)
black point (the "0", or average of all dark pixels): 512
white point (brightest pixels): 16383
read noise (standard deviation of the dark pixels): 2.80
90D DR at 32MP: log2((16383 - 512) / 2.80) = 12.47 EV
normalization 32MP -> 8MP: log2(sqrt(32/8)) = 1.0 EV
90D DR at 8MP: 12.47 + 1.0 = 13.47
This is the "engineering" DR. Or just DR, because PDR is a less known method published by a highly respectable but single site on the Internet IIRC (PDR uses another definition, essentially shifting up the lower bound to some more "photographically aceptable" value taking into account the sensor SNR at low signal levels).
In sensors which have no masked pixels (as the Sony), we need to shoot a 1/8000 frame with the lens cap on to measure the read noise. Another sensors not even have the 0 shifted (don't add e.g. 512) so it is not possible to calculate the standard deviation because the negative signal is lost (the only way it would be measuring patches with diminishing exposure and build a model to predict the lower limit).
It is recommendable to use the green channel because Canon has joined to the manufacturers scaling the other colors (R and B) to apply some kind of white balance "normalization" to the RAW data.
The dcraw tool used to get the raw data don't works with CR3 files, so you need to convert them with the Adobe DNG Converter and use the DNG file, which fortunately includes also the masked pixels (I tested it with CR2 files and I'm sure that DNG doesn't alters the raw data). There is a RAW library out there and some day I'll try it, because dcraw seems dead.
EDIT. Clarkvision has only a minor flaw in my opinion: the author likes to compare the maximum signal level at ISO 100 (full well) with the read noise at higher ISO, which decreases (measured in electrons) as ISO increases. So it gets astounding DR measures for Canon sensors in his tables, as good or better than other manufacturers. IMHO we should compare both values coming from the same ISO level (as everybody else does).
PDR is a less known method published by a highly respectable but single site on the Internet IIRC (PDR uses another definition, essentially shifting up the lower bound to some more "photographically acceptable" value taking into account the sensor SNR at low signal levels).
Jim Kasson frequently computes PDR but less often recently as he always gets the same results
PDR uses a different lower bound but it's not simply shifted.
8MP normalized Engineering Dynamic Range (EDR) (DxOMark Landscape Dynamic Range) is based entirely on Read Noise in DN and resolution.
PDR is based on the Photon Transfer Curve (PTC) and a particular resolution depended Signal to Noise Ratio (SNR).
PTC is a function of Read Noise and gain (also PRNU but that's no relevant to PDR at such low Signal values) .
Gain relates directly to Full Well Capacity (FWC), so PDR is influenced by FWC.
Finally, PDR is measured from an actual test image so other sources of noise, such as Fixed Pattern Noise (FPN), could influence the result.
Tech-savvy people interested in this topic might want to examine the PTC for the 80D.
On this PTC the black dot indicates where PDR is located.
Camperjim wrote:
Where did you come up with this concern? Did you read any of the preliminary reviews? It is clear the 90D has a strong AF system.
I don't see a concern about DR to be overblown. It is an important factor for most of us. Fortunately it seems to be substantially improved.
The predominant focus on this camera appears to be DR and while that is important, I can’t seem to remember when Canon released a successor with significantly worse DR. For my purposes, wildlife and sports, AF is key and the sad truth is that Canon’s ability in this area suffers mightily compared to Nikon and Sony. FWIW, I have a beautiful series of high DR oof, white blobs shot with my 5dmkiv and 500 f4 is ii this weekend which appeared as a tack sharp egret on my buddy’s D500 with the 500 pf. Same conditions, similar settings, and in fact I noticed and focused in first, calling out the flight. His rig $5k, mine more than $11k.
That aside, it is not clear that the 90d af is “strong” and the lack of a spot af is disturbing.
armd wrote:
For my purposes, wildlife and sports, AF is key and the sad truth is that Canon’s ability in this area suffers mightily compared to Nikon and Sony. FWIW, I have a beautiful series of high DR oof, white blobs shot with my 5dmkiv and 500 f4 is ii this weekend which appeared as a tack sharp egret on my buddy’s D500 with the 500 pf. Same conditions, similar settings, and in fact I noticed and focused in first, calling out the flight. His rig $5k, mine more than $11k.
That aside, it is not clear that the 90d af is “strong” and the lack of a spot af is disturbing....Show more →
Sounds like a problem with your gear (or certain combinations of bodies/lenses) and not widespread to all copies. Many of us shoot the same things with no issues, but with different lenses like the 100-400 II with a 1.4x or 2x, ditto with the 70-200 II, and other lenses.
I think most 80D users would like to have better AF performance. I think part of the reason the dialogue is so DR oriented is that it is easier to quantify from the sample pics the (potential) improvements. Quantifying AF performance upgrade is a bit tougher to do from the sample pics to date.
Folks will likely have to get 'em in their hands and make their own assessment if it is indeed improved, or just kissed with something to slap an "improved" label on its AF performance. FPS & DR can be an easier thing to quantify.
armd wrote:
The predominant focus on this camera appears to be DR and while that is important, I can’t seem to remember when Canon released a successor with significantly worse DR. For my purposes, wildlife and sports, AF is key and the sad truth is that Canon’s ability in this area suffers mightily compared to Nikon and Sony. FWIW, I have a beautiful series of high DR oof, white blobs shot with my 5dmkiv and 500 f4 is ii this weekend which appeared as a tack sharp egret on my buddy’s D500 with the 500 pf. Same conditions, similar settings, and in fact I noticed and focused in first, calling out the flight. His rig $5k, mine more than $11k.
That aside, it is not clear that the 90d af is “strong” and the lack of a spot af is disturbing....Show more →
Egrets represent little challenge for combinations much further down the totem pole that the 5D4 and 500LII. The only reason for the performance you claim is user or equipment error. I suggest you have your gear looked at by Canon. I use the same combo for much faster and smaller species and maintain a vey high keeper rate. Handheld. Out of curiosity can I ask if your concern with the 90D having spot AF is because you use this AF selection for BIF?
I see, so it's using the optical black data. That's interesting. I didn't know that RAW file also includes optical black.
Since the method looks solid, my suggestion is to also test the dynamic range in silent shutter/burst mode to check if there is also a fast-but-noisy mode like some modern Sony sensors (3rd gen A7 and Fuji XT3)
vtn54 wrote:
Thanks For Info, But the RAW files in DPReview gallery all are CRAW (Lossy Compressed RAW Files) and DR of these files will be different from true RAW files without lossy compression (but I don't know how much). https://www.photographyblog.com/reviews/canon_eos_90d_review/preview_images
No difference when I tried ISO 100 and ISO 6400 with Hraw.
FWIW Bill Claff's data does not appear to support this
The full PDR graph is not yet available for the 90D but presumably it will reflect what we see in the read noise:
1. Base ISO will be similar to 80D
2. High ISO performance will be closer to 7D2 or slightly better
mb126 wrote:
FWIW Bill Claff's data does not appear to support this
The full PDR graph is not yet available for the 90D but presumably it will reflect what we see in the read noise:
1. Base ISO will be similar to 80D
2. High ISO performance will be closer to 7D2 or slightly better
I think that is basically the same good result? Similar DR to the 80D at pixel level, hence a slight improvement normalized. High ISO read noise similar to the 7D2 is also very good, as is was always better than the 80D. Normalized it will be better, and the best Canon APS-C ever?
What is it that Bill Claff's data doesn't support?
cgarcia wrote:
Canon has refined the read noise performance (not yet to the best of the industry) which is what matters for DR and shadows performance. Photon noise sets an unavoidable upper limit in the SNR (signal-to-noise ratio) under good lighting and modern sensors are all very similar there because have similar efficiency.
I haven't look since a lot of time to APS-C sensors (the 90D is a exception due to my Canon roots) but I think that Canon has not yet catched Nikon or Sony APS-C flagships DR (IIRC some of these went beyond 14 EV a long time ago). But at least now has decent figures. ...Show more →
Check Bill Claff's photonstophotos site: http://www.photonstophotos.net/Charts/PDR.htm#Canon%20EOS%207D%20Mark%20II,Canon%20EOS%2080D,Nikon%20D500,Nikon%20D7200
The 80D and 7DII are very similar to the top Nikons at high isos. And, remember the Nikons have a 1.5 crop factor that gives them a few % advantage over Canon's 1.6 since Bill normalises the PDR to an 8x10" print, which means that the Canons have to be blown up more. All of the top APS-Cs have very similar PDRs at high iso, which makes your results very surprising.
alundeb wrote:
I think that is basically the same good result? Similar DR to the 80D at pixel level, hence a slight improvement normalized. High ISO read noise similar to the 7D2 is also very good, as is was always better than the 80D. Normalized it will be better, and the best Canon APS-C ever?
What is it that Bill Claff's data doesn't support?
Based on magnitude of difference between 80D vs 90D and 80D vs A6300, I don't think we are likely to see anywhere near a 1 stop improvement as OP suggests.
) but I think that Canon has not yet catched Nikon or Sony APS-C flagships DR (IIRC some of these went beyond 14 EV a long time ago). But at least now has decent figures.
