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

Archive 2015No DR improvement = no 5Ds for me

  
 
snapsy
Offline
Details
p.15 #1No DR improvement = no 5Ds for me
RustyBug wrote:
snapsy wroteAs indicated before (here), the midtones between high MP and lower MP sensors of the same technology will have nearly identical noise since the midtones are dominated by shot noise, which is a function of QE. Where the difference occurs is in the shadows at high ISO, where the higher MP sensor's pixel count results in a higher aggregate read noise, which translates into lower High ISO DR, which is a proxy for shadow noise/read noise.
Driving the (lay observation) correlation of smaller pixels = more noise.

snapsy wroteBtw Rusty, the dpreview samples you posted don't appear to have had the 'print' tab selected, so you were also comparing them at their native pixel sizes rather than normalized to sensor area. To see how much of a difference that makes, here are 100% crops of a D3s vs D800 High ISO comparison I did in 2012 at ISO 12,800, first at their native pixel sizes and then normalized to 12MP. Notice again how the midtones are almost identical and the shadow favors the D3s.
+1 @ the normalization.
The story repeats @ detectability to the low end. I also notice a color variation between the D3s and the D800 ... the D800 showing a less neutral (i.e. more blue) rendering in the camera body / lens.


The lay observation once thought the earth to be flat but applying a bit more nuanced thought to the observation revealed otherwise. The D800 has a shadow cast due to a combination of amp noise and lower High ISO DR.
Feb 10, 2015 at 08:56 AM
snapsy
Offline
Details
p.15 #2No DR improvement = no 5Ds for me
RustyBug wrote:
takowasa wroteCanon sensors fare well at high ISOs (same quantum efficiency as everyone else, and the same, or even a tad lower, electronic noise per pixel). The "deficiencies" of Canon sensors compared to the competition are low pixel count and high electronic noise at low ISOs.

Lower electronic noise @ high ISO vs. high electronic noise low ISO

Sounds like there is a crossover that occurs ... diff mfr's opting to lean one way vs. the other.

Some are forced to lean more than others due to limitations in their technology investment. The Sony A7s has lower read noise at low ISO than most traditional sensors and the lowest read noise at high ISO than any sensor.
Feb 10, 2015 at 08:59 AM
RustyBug
Offline
Details
p.15 #3No DR improvement = no 5Ds for me
snapsy wrote:

The D800 has a shadow cast due to a combination of amp noise and lower High ISO DR.


amp noise

I tend to like my blacks, black.


snapsy wrote:
Some are forced to lean more than others due to limitations in their technology investment. The Sony A7s has lower read noise at low ISO than most traditional sensors and the lowest read noise at high ISO than any sensor.


This seems to be an improvement over their own previous efforts.
Feb 10, 2015 at 09:11 AM
snapsy
Offline
Details
p.15 #4No DR improvement = no 5Ds for me
RustyBug wrote:
snapsy wroteSome are forced to lean more than others due to limitations in their technology investment. The Sony A7s has lower read noise at low ISO than most traditional sensors and the lowest read noise at high ISO than any sensor.
This seems to be an improvement over their own previous efforts.


Before the A7s Sony didn't have a compelling reason to develop a low MP version of the FF Exmor sensor. Their biggest customer (Nikon) already had their own and the volumes are also much lower. But for the A7s Sony needed a FF sensor with a fast enough readout to support full sampling of the sensor for 1080 and 4K video and that required backing off on the pixel count.
Feb 10, 2015 at 09:20 AM
RustyBug
Offline
Details
p.15 #5No DR improvement = no 5Ds for me
So, while their motive for backing off on the pixel count may have been video ... it's benefits of the larger pixels are realized anyway. Is that correct?
Feb 10, 2015 at 09:22 AM
snapsy
Offline
Details
p.15 #6No DR improvement = no 5Ds for me
RustyBug wrote:
snapsy wroteThe D800 has a shadow cast due to a combination of amp noise and lower High ISO DR.
amp noise

I tend to like my blacks, black.

The amp noise was significantly reduced in the 24MP D750 Exmor and I imagine will continue to improve in future generations. Some speculate the noise is from the high density of on-chip logic to support the Exmor's massively parallel readout structure. More info here.
Feb 10, 2015 at 09:24 AM
snapsy
Offline
Details
p.15 #7No DR improvement = no 5Ds for me
RustyBug wrote:
So, while their motive for backing off on the pixel count may have been video ... it's benefits of the larger pixels are realized anyway. Is that correct?


Yes but the one benefit of large pixels (lower aggregate readout noise @ high ISO) is a temporary situation from limitations in current technology. It will likely be solved in the intermediate future. There are other sensor designs that have already bridged the gap significantly, like the Toshiba-designed D7100 APS-C sensor.
Feb 10, 2015 at 09:29 AM
RustyBug
Offline
Details
p.15 #8No DR improvement = no 5Ds for me
snapsy wrote:

like the Toshiba-designed D7100 APS-C sensor.


Suggesting the former comparison using the D7100 wasn't really an apples to apples one on the matter of pixel size @ non- ceteris paribus electronics.

Variables, variables and more variables.
Feb 10, 2015 at 09:35 AM
snapsy
Offline
Details
p.15 #9No DR improvement = no 5Ds for me
RustyBug wrote:
snapsy wrotelike the Toshiba-designed D7100 APS-C sensor.
Suggesting the former comparison using the D7100 wasn't really an apples to apples one on the matter of pixel size @ non- ceteris paribus electronics.


The Toshiba sensor has the highest pixel density of any APS-C or FF sensor (at least before the NX1) yet has the lowest readout noise of any APS-C sensor and even lower than most FF sensors. Since sensors don't come out every day you have to cull the technology information and trends from what's available.
Feb 10, 2015 at 09:38 AM
RustyBug
Offline
Details
p.15 #10No DR improvement = no 5Ds for me
Understood @ "what's available" ... just a believer in understand the diff's @ variables in conjunction with making assessments.

+1 @ tough gig sometimes for consumers outside the lab and engineer's buildings.


I found some spec on a camera given in these terms. Not sure how to convert them to meaningful information. Can someone convert that to DR or noise or or relate it somehow something a bit more meaningful.

65.4dB dynamic range, 65,000 electron well-capacity, 36x24, 14MP
Feb 10, 2015 at 09:58 AM
takowasa
Offline
Details
p.15 #11No DR improvement = no 5Ds for me
RustyBug wrote:
Understood @ "what's available" ... just a believer in understand the diff's @ variables in conjunction with making assessments.

+1 @ tough gig sometimes for consumers outside the lab and engineer's buildings.

I found some spec on a camera given in these terms. Not sure how to convert them to meaningful information. Can someone convert that to DR or noise or or relate it somehow something a bit more meaningful.

65.4dB dynamic range, 65,000 electron well-capacity, 36x24, 14MP


6 dB corresponds to 1 stop of DR, so that's just shy of 11 stops of per-pixel engineering DR (65.4 / 6 ~ 10.86 ev). Combine that with the 65000 electron FWC and we get a base ISO electronic noise of 65000 / 2^10.86 = 35 electrons.
Feb 10, 2015 at 07:34 PM
RustyBug
Offline
Details
p.15 #12No DR improvement = no 5Ds for me
Okay, thanks.

Diff line of question @ ISO.

I went to sensorgen.info (per the provided link elsewhere). I compiled and tabulated all the FF sensors they had listed.

I compared the stated ISO to the measured ISO and derived the % of the two. As an average by mfr, the Canon sensors averaged slightly above 80% of stated ISO while Nikon & Sony averaged around 75% of stated ISO (restricted to the base ISO + 3 stops). If we include the -1 stop @ ISO 100 for Nikon/Sony, Nikon came up to match Canon around 80% @ ranges of 100-1600).

Going above 1600, they all were very close @ 72%, 73%, 74%.

My question is ... to what degree does our stated vs. actual variance influence our expectation / assessment of noise or DR, etc.? It just strikes me that this variance has a factor in things ... just unsure of what utility it might be for us when calculations are typically made @ stated values vs. actual values. Splittin' hairs, maybe ... maybe not @ it offers something for why we hear about Sony's "lead" on things. I mean, if they are shaving 5% ... just wondering if that accounts for any degree of variance (i.e. @ more light vs. less light vs. shot noise)?

I've also got the read noise, saturation, pixel size, pixel count, etc. for additional look into things. The preliminary review seems to be able to reveal a "sweet spot" in each camera where you get the most ISO for the least noise and least loss of DR ... still working toward that, but noting the ISO variance, I thought I might want to understand it's impact a bit better first.

Feb 11, 2015 at 09:06 AM
takowasa
Offline
Details
p.15 #13No DR improvement = no 5Ds for me
RustyBug wrote:
Okay, thanks.

Diff line of question @ ISO.

I went to sensorgen.info (per the provided link elsewhere). I compiled and tabulated all the FF sensors they had listed.

I compared the stated ISO to the measured ISO and derived the % of the two. As an average by mfr, the Canon sensors averaged slightly above 80% of stated ISO while Nikon & Sony averaged around 75% of stated ISO (restricted to the base ISO + 3 stops). If we include the -1 stop @ ISO 100 for Nikon/Sony, Nikon came up to match Canon around 80% @ ranges of 100-1600).

Going above 1600,
...Show more →

I can't speak for peoples' expectations, but in terms of noise, it has no bearing. Someone could make a camera and label the ISOs "A", "B", "C", etc., and, in terms of noise, it wouldn't make any difference whatsoever.

It just strikes me that this variance has a factor in things ... just unsure of what utility it might be for us when calculations are typically made @ stated values vs. actual values. Splittin' hairs, maybe ... maybe not @ it offers something for why we hear about Sony's "lead" on things. I mean, if they are shaving 5% ... just wondering if that accounts for any degree of variance (i.e. @ more light vs. less light vs. shot noise)?

I've also got the read noise, saturation, pixel size, pixel count, etc. for additional look into things. The preliminary review
...Show more →

The more light you get on the sensor, the less noise the photo will have. To get more light on the sensor, however, you need a longer exposure time, which can result in motion blur or camera shake in lower light, or oversaturation in good light. Alternatively, you can also use a wider aperture for a given exposure time, but this will result in a more shallow DOF and, depending on the lens and how wide you go, greater lens aberrations.
Feb 11, 2015 at 12:32 PM
RustyBug
Offline
Details
p.15 #14No DR improvement = no 5Ds for me
While it isn't a perfect 1:1 correlation, there does seem to be an fairly strong inverse relationship between the lower measured ISO and the increased DR @ things that make you go hmmmm. Just something that caught my eye, even though I don't know the reason why.


ISO MISO % Noise Saturation DR
100 92 92% 63.7 102167 10.6
100 89 89% 49.6 090877 10.8
100 84 84% 41.2 061268 10.5
100 80 80% 38.5 090101 11.2
100 80 80% 26.4 074256 11.5
100 80 80% 33.6 068151 11.0
100 79 79% 07.4 076444 13.3
100 78 78% 06.0 047765 13.0
100 75 75% 14.0 125597 13.1
100 75 75% 18.8 118339 12.6
100 75 75% 04.5 049554 13.4
100 74 74% 21.7 134567 12.6
100 74 74% 05.7 074971 13.7
100 74 74% 04.6 048818 13.4
100 73 73% 05.5 081608 13.9
100 73 73% 05.2 054924 13.4

It would seem to me that if you normalized for the measured ISO, the amount of light actually reaching the sensor would change ... and that would then influence the noise (@ more light = less noise > less noise = more DR) for a normalized ISO rather than a reported one. Curious at the math between the extremes @ ISO 92 vs. ISO 73 and 13.4 vs. 10.6 ... they are both a factor of 126%.

Edited on Feb 11, 2015 at 01:30 PM · View previous versions
Feb 11, 2015 at 01:01 PM
takowasa
Offline
Details
p.15 #15No DR improvement = no 5Ds for me
RustyBug wrote:
While it isn't a perfect 1:1 correlation, there does seem to be an fairly strong inverse relationship between the lower measured ISO and the increased DR @ things that make you go hmmmm. Just something that caught my eye, even though I don't know the reason why.

ISO MISO % Noise Saturation DR
100 92 92% 63.7 102167 10.6
100 89 89% 49.6 090877 10.8
100 84 84% 41.2 061268 10.5
100 80 80% 38.5 090101 11.2
100 80 80% 26.4 074256 11.5
100 80 80% 33.6 068151 11.0
100 79 79% 07.4 076444 13.3
100 78 78% 06.0 047765 13.0
100 75 75% 14.0 125597 13.1
100 75 75% 18.8 118339 12.6
100 75 75% 04.5 049554 13.4
100 74 74% 21.7 134567 12.6
100 74 74% 05.7 074971 13.7
100 74 74% 04.6 048818 13.4
100 73 73% 05.5 081608 13.9
100 73 73% 05.2 054924 13.4


The ISO setting, either nominal or measured, has no bearing on the DR. The [engineering] DR is simply the number of stops from the electronic (read) noise to the saturation point. Typically, however, each stop we raise the ISO setting halves (more or less) the saturation point, which would also reduce the DR by a stop if the electronic noise stayed the same.

All that said, the DR doesn't tell the story about noise in the photo. If anything, it's more an indicator of shadow noise than anything else. For more details, see here:

http://www.josephjamesphotography.com/equivalence/#dr
Feb 11, 2015 at 01:21 PM
RustyBug
Offline
Details
p.15 #16No DR improvement = no 5Ds for me
takowasa wrote:
http://www.josephjamesphotography.com/equivalence/#dr


That'll be a quick read.
Feb 11, 2015 at 01:52 PM
takowasa
Offline
Details
p.15 #17No DR improvement = no 5Ds for me
RustyBug wrote:
That'll be a quick read.


Time's up.
Feb 11, 2015 at 04:57 PM
Jeff Nolten
Offline
Details
p.15 #18No DR improvement = no 5Ds for me
It was a good read too, the parts I read. I liked the noise section too. Thanks for posting the reference.
Feb 11, 2015 at 05:07 PM
FM ForumsCanon ForumJoin Upload & Sell
  
 

You are not logged inLog inRegister