fredmiranda.com Login
This topic is in the 2015 archive. You can read it and like its posts here. It is read only.

Archive 2015Canon 5DS dynamic range: zero improvement (RAWs analyzed)

  
 
ggreene
Offline
Details
p.2 #1Canon 5DS dynamic range: zero improvement (RAWs analyzed)
retrofocus wrote:
Canon police early in action . Trying to rule out every sort of criticism or cons. No surprises here.


Looks like the DRones were there first though. Apparently, the Sony and Nikon forums aren't entertaining enough to keep them occupied.

Sales will tell the tale as to the direction of Canon priorities. So far low ISO DR hasn't really helped Nikon or Sony sell camera's. If not for the Apple contract with iPhones even Sony's Exmor sensor would be struggling. Their own forecasting predicts that even with the Exmor sensors they can't stop the decline.

Mar 22, 2015 at 02:13 PM
alundeb
Offline
Details
p.2 #2Canon 5DS dynamic range: zero improvement (RAWs analyzed)
cgarcia wrote:
These results should be readed with slight caution. Dcraw didn't show a black point. By visual inspection of the data neither I have been able to infere it. What is even more disturbing is that exiftool doesn't found the black level tags stored in the EXIF information. A potential explanation is that Canon may have removed it: let's hope this is not the case, since this was one of the good things about Canon RAWs (and that would make very, very unhappy to astrophotograpy folks). Hope that this is due to some strange side-effect at decoding. Also note that the
...Show more →

A histogram of the read noise values would quickly answer that. Is it possible for youy to provide that without too much effort? Thank you!
Mar 22, 2015 at 02:33 PM
alundeb
Offline
Details
p.2 #3Canon 5DS dynamic range: zero improvement (RAWs analyzed)
The "best case" interpretation of the data is 1/4 stops improved pixel level DR at ISO 100 over the 7D II for raw data. If this is combined with a 1/3 stops denser CFA, the effect will be slightly more for a converted color image because less amplicifaction of the color channels is needed.

If we then combine this with the anticipated absence of banding similar to the 7DII, and the high pixel count that increases normalized DR, we get a substantial improvement in usable DR over the 5D III.

(Not excusing Canon, just trying to get accurate data).
Mar 22, 2015 at 02:44 PM
memoria
Offline
Details
p.2 #4Canon 5DS dynamic range: zero improvement (RAWs analyzed)
Which software do you use when editing the 5Ds RAW files?
Mar 22, 2015 at 03:32 PM
cgarcia
Offline
Details
p.2 #5Canon 5DS dynamic range: zero improvement (RAWs analyzed)
alundeb wrote:
cgarcia wroteThese results should be readed with slight caution. Dcraw didn't show a black point. By visual inspection of the data neither I have been able to infere it. What is even more disturbing is that exiftool doesn't found the black level tags stored in the EXIF information. A potential explanation is that Canon may have removed it: let's hope this is not the case, since this was one of the good things about Canon RAWs (and that would make very, very unhappy to astrophotograpy folks). Hope that this is due to some strange side-effect at decoding. Also note that the estimated DR would be about 0.2 stops less at ISO 100 in case the black level (not substracted in the above calculations) were 2048.
A histogram of the read noise values would quickly answer that. Is it possible for youy to provide that without too much effort? Thank you!


Yes, you have reason!. Appended at the end of this post for ISO 100. It clearly points to about 2048. By calculating the stats:

ISO 100 masked pixels: mean=2047.48 min=2015 max=2118
ISO 3200 masked pixels: mean=2047.4 min=1643 max=2468

The 6D raw files have just a 2047 black point value. I assume that exiftool has failed to report the two related fields (despite likely stored in the EXIF metadata, not reverse-engineered from the raw).

Using that to correct the previous estimation:

ISO 0100 - DR 12.30 at 8MP (10.99 at 50MP) read noise 6.31819 DN (start 6.23) saturation 12686 DN
ISO 0200 - DR 12.30 at 8MP (10.98 at 50MP) read noise 6.62953 DN (start 5.98) saturation 13336 DN
ISO 0400 - DR 12.08 at 8MP (10.76 at 50MP) read noise 7.71204 DN (start 6.12) saturation 13336 DN
ISO 0800 - DR 11.66 at 8MP (10.34 at 50MP) read noise 10.3258 DN (start 5.93) saturation 13337 DN
ISO 1600 - DR 10.98 at 8MP (9.66 at 50MP) read noise 16.5492 DN (start 6.06) saturation 13338 DN
ISO 3200 - DR 10.17 at 8MP (8.85 at 50MP) read noise 29.079 DN (start 6.83) saturation 13339 DN
ISO 6400 - DR 9.38 at 8MP (8.06 at 50MP) read noise 50.119 DN (start 9.23 saturation 13342 DN
ISO 12800 - DR 8.41 at 8MP (7.09 at 50MP) read noise 98.2089 DN (start 18.65) saturation 13348 DN


ISO 100 read noise histogram (0.5 million pixels sample)
Data number - count of masked pixels
2015 1
2019 1
2020 1
2021 1
2022 2
2023 14
2024 12
2025 38
2026 17
2027 80
2028 127
2029 350
2030 642
2031 166
2032 827
2033 1940
2034 3368
2035 3710
2036 947
2037 7320
2038 11559
2039 11948
2040 20023
2041 733
2042 27171
2043 24335
2044 36907
2045 41002
2046 4
2047 32056
2048 42900
2049 40809
2050 36867
2051 2094
2052 22083
2053 26832
2054 20631
2055 3616
2056 9426
2057 10707
2058 7734
2059 3940
2060 2927
2061 686
2062 1737
2063 850
2064 687
2065 329
2066 59
2067 135
2068 111
2069 66
2070 6
2071 43
2072 66
2073 80
2074 47
2075 15
2076 96
2077 122
2078 117
2079 115
2080 142
2081 69
2082 210
2083 167
2084 175
2085 62
2086 162
2087 149
2088 175
2089 122
2090 107
2091 56
2092 106
2093 88
2094 55
2095 60
2096 21
2097 52
2098 38
2099 35
2100 18
2101 11
2102 18
2103 5
2104 5
2105 6
2106 4
2107 1
2108 1
2109 2
2110 2
2118 1


Mar 22, 2015 at 04:49 PM
cgarcia
Offline
Details
p.2 #6Canon 5DS dynamic range: zero improvement (RAWs analyzed)
memoria wrote:
Which software do you use when editing the 5Ds RAW files?


dcraw with the options"-D -4 -j -t 0 -v"

On this way, we have access to the full raw data capture (14-bit pixel values stored in 16 bit integers). This includes the image borders covered to not capture light (useful to measure the sensor read noise).

The decoding is wrong (a 17792x2960 image). The left half seems to have good data, though.Usually dcraw is able to decode future sensor images (except for fine color adjustments) but this is not the case for the 5DS. We'll have to wait a bit.

The vertical masked area ends stored in two places in this weird geometry:

x=0, y=64, width=160, height=2896
x=4448, y=64, width=160, height=2896

To calculate stats I use a small custom made C++ application I started a few months ago. I'll publish it for free one of these days if I finally end it (currently requires manually editing the coordinates).
Mar 22, 2015 at 05:05 PM
chez
Offline
Details
p.2 #7Canon 5DS dynamic range: zero improvement (RAWs analyzed)
alundeb wrote:
cgarcia, thank you for the analysis!

The saturation vaule at ISO 100 compared to the 7D II indicates either 1/3 stops higher Full Well Capacity or 1/3 stops denser Color Filter Array (most likely), or a combination.


Yes, thank you very much...even though some of the regulars jump in and down play your work. I for one appreciate it very much. It's just another datum I use in determining if these new cameras are for me. Don't understand some of the other posters getting into a defensive mood. If they don't like your approach and results, then just don't use your findings...but please don't put down someone that has done some hard scientific work...at least unless you have your own analysis which contradicts these findings. If so...put up or you know what.
Mar 22, 2015 at 05:59 PM
Monito
Offline
Details
p.2 #8Canon 5DS dynamic range: zero improvement (RAWs analyzed)
cgarcia wrote:
The decoding is wrong (a 17792x2960 image).


Thanks, but I think I will wait for more careful rigorous analysis with fully equipped decoders before taking it on board as a data point.

All the same, for my way of working (EttR, Canon EF mounts), I am likely to buy a 5Ds if I can swing it, with all its limitations and advantages. I have no doubt that, as with all previous cameras, I will continue to be held back more by my own limitations than the camera's.
Mar 22, 2015 at 06:04 PM
snapsy
Offline
Details
p.2 #9Canon 5DS dynamic range: zero improvement (RAWs analyzed)
Monito wrote:
cgarcia wroteThe decoding is wrong (a 17792x2960 image).
Thanks, but I think I will wait for more careful rigorous analysis with fully equipped decoders before taking it on board as a data point.

All the same, for my way of working (EttR, Canon EF mounts), I am likely to buy a 5Ds if I can swing it, with all its limitations and advantages. I have no doubt that, as with all previous cameras, I will continue to be held back more by my own limitations than the camera's.


I agree about ETTR being a viable strategy to ameliorate DR issues in many shooting scenarios. That plus future ML support with advanced bracketing. If I can work around the 5D original's DR and banding then it's certainly doable on a body with even better DR and no banding. I don't have plans for the 5DS myself though.
Mar 22, 2015 at 06:33 PM
StillFingerz
Offline
Details
p.2 #10Canon 5DS dynamic range: zero improvement (RAWs analyzed)
cgarcia wrote:
memoria wroteWhich software do you use when editing the 5Ds RAW files?
dcraw with the options"-D -4 -j -t 0 -v"

On this way, we have access to the full raw data capture (14-bit pixel values stored in 16 bit integers). This includes the image borders covered to not capture light (useful to measure the sensor read noise).

The decoding is wrong (a 17792x2960 image). The left half seems to have good data, though.Usually dcraw is able to decode future sensor images (except for fine color adjustments) but this is not the case for the 5DS. We'll have to wait a bit.

The vertical masked area ends stored in two places in this weird geometry:

x=0,
...Show more →

Since your calcs, the software and the design unknowns of a 'new' sensor and it's RAW files that are pre-production are only in limbo; could change at any moment, like when production RAWs are available, how can you declare 'zero' change in DR, you can't.

That said the math and software 'Dcraw' seem an interesting concept/methodology for evaluating sensors, yet it's nowhere conclusive, so all you've done is report inconclusive data with a methodology that has reportedly work before on 'production' targets, doing so on a non-production target sample...

This smells of sensationalism, guessing if there will be a double yoke before cracking the egg kind if thing. Much like what the media does with their 'story telling' disguised as 'news', often causing damage before the 'real' truth is revealed. In this case, I'll hazard a wild arse guess that 'ZERO' doesn't equal 'Zero' when the final numbers are derived, and thus your 'headline' was a false statement; until proven correct, with zero doubt.

On another note, as a C++ coder myself; began in the mid-90s, I'd be most interested in your application and it's implementation...I'm always looking to learn a new trick or two, or three...no joking

Cheers,
Jerry
Mar 22, 2015 at 08:28 PM
chez
Offline
Details
p.2 #11Canon 5DS dynamic range: zero improvement (RAWs analyzed)
StillFingerz wrote:
cgarcia wrote
memoria wroteWhich software do you use when editing the 5Ds RAW files?
dcraw with the options"-D -4 -j -t 0 -v"

On this way, we have access to the full raw data capture (14-bit pixel values stored in 16 bit integers). This includes the image borders covered to not capture light (useful to measure the sensor read noise).

The decoding is wrong (a 17792x2960 image). The left half seems to have good data, though.Usually dcraw is able to decode future sensor images (except for fine color adjustments) but this is not the case for the 5DS. We'll have to wait a bit.

The vertical masked area ends stored in two places in this weird geometry:

x=0, y=64, width=160, height=2896
x=4448, y=64, width=160, height=2896

To calculate stats I use a small custom made C++ application I started a few months ago. I'll publish it for free one of these days if I finally end it (currently requires manually editing the coordinates).
Since your calcs, the software and the design unknowns of a 'new' sensor and it's RAW files that are pre-production are only in limbo; could change at any moment, like when production RAWs are available, how can you declare 'zero' change in DR, you can't.

That said the math and software 'Dcraw' seem an interesting concept/methodology for evaluating sensors, yet it's nowhere conclusive, so all you've done is report inconclusive data with a methodology that has reportedly work before on 'production' targets, doing so on a non-production target sample...

This smells of sensationalism, guessing if there will be a double yoke
...Show more →

Jerry, until we have a release camera, we will never know the actual performance...but given the data we have today, this information is very valid. Sure the sensor performanc can and will most likey change...but given what we have today, this is very relevant data.

Love to see the exact analysis done on a production version.

By the way...I would think the sensor is very close to production version if it has been released by Canon for pre-release testing.
Mar 22, 2015 at 08:50 PM
Jeffrey
Offline
Details
p.2 #12Canon 5DS dynamic range: zero improvement (RAWs analyzed)
Why do people still start these useless threads? Yeah, I know, If I don't like them then why do I read them... Duh, because I thought there might have been some pertinent interest for me but there is none. 11 stops of DR is just fine by me. The rest is the same recycled fodder for bored camera owners. Sorry to be brazenly straightforward with my opinions. Oh, not really!
Mar 22, 2015 at 09:13 PM
arbitrage
Offline
Details
p.2 #13Canon 5DS dynamic range: zero improvement (RAWs analyzed)
If Canon had made a higher DR sensor I'm sure it would have made the marketing materials. I think this test is a valid data point and confirms what we already knew and what Canon already told us. Anyone who really thinks that in 3 months this sensor is some how going to be radically different then it is in current demo models is kidding themselves IMHO. If you need DR now buy the D810 or the Sony A7 whatever. If you need 50MP and DR then Sony will soon have that also and Nikon will likely be not to far behind. I wouldn't doubt the Sony 50MP Exmor will be out on store shelves before the 5Ds is.
Mar 22, 2015 at 09:27 PM
zlatko
Offline
Details
p.2 #14Canon 5DS dynamic range: zero improvement (RAWs analyzed)
retrofocus wrote:
Canon police early in action . Trying to rule out every sort of criticism or cons. No surprises here.


Constant repetition of alleged DR deficiencies. What else is new? It would be the same as posting frequent reminders on Sony / Nikon forums of the various lenses their systems lack. Or droning on about their ergonomic deficiencies. How would they receive that information?

Edited on Mar 22, 2015 at 10:11 PM · View previous versions
Mar 22, 2015 at 10:05 PM
zlatko
Offline
Details
p.2 #15Canon 5DS dynamic range: zero improvement (RAWs analyzed)
cgarcia wrote:
It seems that the 5DS has yet another boring Canon sensor with, again, the same boring 11 f-stop DR at pixel level on base ISO.


This sort of analysis is useful for showing that if one couldn't make decent photos with past Canon gear, then one won't be able to make decent photos with the new Canon gear either. There is no such thing as boring sensors, only boring photographers who obsess over DR at pixel level on base ISO and other such things.
Mar 22, 2015 at 10:09 PM
mttran
Offline
Details
p.2 #16Canon 5DS dynamic range: zero improvement (RAWs analyzed)
arbitrage wrote:
If Canon had made a higher DR sensor I'm sure it would have made the marketing materials. I think this test is a valid data point and confirms what we already knew and what Canon already told us. Anyone who really thinks that in 3 months this sensor is some how going to be radically different then it is in current demo models is kidding themselves IMHO. If you need DR now buy the D810 or the Sony A7 whatever. If you need 50MP and DR then Sony will soon have that also and Nikon will likely be not to
...Show more →

+1, Cgarcia just confirmed what canon has said that 5Dsr DR is about the same as 5D2. Question now is any shadows banding when hitting the max DR range, we will find out soon.
Mar 22, 2015 at 10:10 PM
gdanmitchell
Offline
Details
p.2 #17Canon 5DS dynamic range: zero improvement (RAWs analyzed)
Just preordered my 5Ds R. :-)
Mar 22, 2015 at 10:38 PM
EB-1
Offline
Details
p.2 #18Canon 5DS dynamic range: zero improvement (RAWs analyzed)
I just hope it is OK at 1600.

EBH
Mar 22, 2015 at 10:39 PM
EB-1
Offline
Details
p.2 #19Canon 5DS dynamic range: zero improvement (RAWs analyzed)
gdanmitchell wrote:
Just preordered my 5Ds R. :-)


I ordered a few copies, too. It's probably too late though.

EBH
Mar 22, 2015 at 10:48 PM
StillFingerz
Offline
Details
p.2 #20Canon 5DS dynamic range: zero improvement (RAWs analyzed)
chez wrote:
StillFingerz wrote
cgarcia wrote
memoria wroteWhich software do you use when editing the 5Ds RAW files?
dcraw with the options"-D -4 -j -t 0 -v"

On this way, we have access to the full raw data capture (14-bit pixel values stored in 16 bit integers). This includes the image borders covered to not capture light (useful to measure the sensor read noise).

The decoding is wrong (a 17792x2960 image). The left half seems to have good data, though.Usually dcraw is able to decode future sensor images (except for fine color adjustments) but this is not the case for the 5DS. We'll have to wait a bit.

The vertical masked area ends stored in two places in this weird geometry:

x=0, y=64, width=160, height=2896
x=4448, y=64, width=160, height=2896

To calculate stats I use a small custom made C++ application I started a few months ago. I'll publish it for free one of these days if I finally end it (currently requires manually editing the coordinates).
Since your calcs, the software and the design unknowns of a 'new' sensor and it's RAW files that are pre-production are only in limbo; could change at any moment, like when production RAWs are available, how can you declare 'zero' change in DR, you can't.

That said the math and software 'Dcraw' seem an interesting concept/methodology for evaluating sensors, yet it's nowhere conclusive, so all you've done is report inconclusive data with a methodology that has reportedly work before on 'production' targets, doing so on a non-production target sample...

This smells of sensationalism, guessing if there will be a double yoke before cracking the egg kind if thing. Much like what the media does with their 'story telling' disguised as 'news', often causing damage before the 'real' truth is revealed. In this case, I'll hazard a wild arse guess that 'ZERO' doesn't equal 'Zero' when the final numbers are derived, and thus your 'headline' was a false statement; until proven correct, with zero doubt.

On another note, as a C++ coder myself; began in the mid-90s, I'd be most interested in your application and it's implementation...I'm always looking to learn a new trick or two, or three...no joking

Cheers,
Jerry
Jerry, until we have a release camera, we will never know the actual performance...but given the data we have today, this information is very valid. Sure the sensor performanc can and will most likey change...but given what we have today, this is very relevant data.

Love to see the exact analysis done on a production version.

By the way...I would think the sensor is very close to production version if it has been released by Canon for pre-release testing.


Mostly agree Chez, I'm not thinking we'll see anymore DR magic in this new mega 35mm sensor design than perhaps the 'cleaner' shadows found in the 6D or the 7D2's, my gut feels Canon isn't to concerned with the DR race with these models. These are very niche bodies, not unlike the 7D/7D2 were/are, serving a smaller group of photographers.

Note also, other than their MP count, which might help me from a cropability standpoint; caused at times by my disability related reach issues, I'm not seeing a buy here. And even though I'm grabbing a 6D for current FF purposes, my eyes are wide open and are following Sony closely.

As mentioned before I'm kind of tied to Canon because only their wired remotes are of a size/design that can be used between my teeth; a big mouth on forums perhaps but physically not, to allow me to hand-hold and shoot. All Sony has to do is make their remote a bit thinner and change the button position, and add a touch screen, and I'd be 'able' to avail myself of their extras, including wider DR. Even in gardens; covered or not, there can be deep shadows filled with much flora that otherwise go unseen.

So I forego the wider DR for now and stay with ergonomics that work with my speshul needs, those that have allowed me the privilege to shoot for near thirty-seven years seated. Hand-holding is one of those last freedoms I'm unwilling to give up. Compromise is a must with a high level spinal injury on so many levels; subject mater just one, and tripods/wheelchairs play not well together

Cheers,
Jerry
Mar 22, 2015 at 10:54 PM
FM ForumsCanon ForumJoin Upload & Sell
  
 

You are not logged inLog inRegister