Panasonic did something special with the Z6 III sensor

Horshack

Forum Pro
Messages
11,231
Solutions
28
Reaction score
12,594
Location
US
Panasonic introduces the S1 II

Panasonic recently released their S1 II body, which incorporates the same 24MP partially-stacked sensor as the Z6 III. As you may have read, the Z6 III represented a noticeable step backwards in low-ISO dynamic range compared to the Z6 II and Z6.

DPR recently posted their S1 II studio samples a few weeks ago and found it had significantly better low-ISO DR vs other cameras including the Z6 III. So good in fact that they wrote:

The S1II provides a good amount of dynamic range, giving you room to pull your shadows up without the detail being overwhelmed by noise. However, given the lack of chroma noise visible in our tests, it seems like there may be some noise reduction going on, either in ACR or in-camera.

Bill posted his S1 II PDR measurements a few weeks ago, which confirmed DPR's findings on improved DR.

Analysis of the S1 II reveals DGO

Initial analysis of the S1 II raw files showed evidence of significant noise filtering, which could be the result of the camera applying raw noise reduction and thus the source of the S1 II's DR "advantage". However, the S1 II's DR was so good that it seemed better than even NR could produce.

I was intrigued, so a few weeks ago I acquired an S1 II and started my own deep technical analysis. That analysis revealed evidence Panasonic was actually employing a dual gain output to achieve the S1 II's excellent DR. This is where the camera performs two readouts of a single exposure, one at ISO 100 and the other ISO 800, then merges the two in the resulting raw. This combines the low ISO saturation capacity of ISO 100 with the deep shadow noise improvement seen at the dual gain step of ISO 800.

I found the S1 II only employs this technique when using the EFCS or fully-mechanical shutter - it does not use it for its electronic shutter.

Two years ago I created a thread demonstrating what the Z8/Z9 sensor would look like if it had DGO (Dual Gain Output). I applied that same simulation to the S1 II's electronic shutter and it produced results identical to the S1 II's mechanical shutter, further demonstrating the camera is using DGO. Finally, I compared blackframe histograms of the S1 II's mechanical shutter to my S1 II electronic shutter DGO sim and found they shared the same scaling anomaly - this IMO represented definitive proof that Panasonic is employing DGO.

Panasonic S1 II vs Z6 III Compared

Here are a series of animations visually comparing the S1 II to the Z6 III. These were shot at ISO 100 with a -7EV exposure, which was adjusted back in post. All are 100% crops with NR disabled.

This first one compares the S1 II's electronic vs mechanical shutter, the later of which has Panasonic's DR-boosting DGO:

Animation: Panasonic S1 II Electronic vs Mechanical Shutter

This second animation compares the S1 II's electronic shutter to the Z6 III. Notice they have identical noise, since the S1 II only employs DGO on its mechanical shutter:

Animation: Panssonic S1 II Electronic vs Z6 II Electronic

Finally, here's an animation comparing the S1 II's DGO mechanical shutter to a simulated Z6 III DGO I created with my code, which combines ISO 100 and 800 exposures in a fashion similar to what Panasonic is likely using:

Animation: Panasonic S1 II Mechanical vs Z6 III Electronic DGO Simulation

Takeaway

Two years ago I simulated the potential noise improvement of dual-gain output applied to a full-frame sensor and hoped we would see it one day in a camera. That day has arrived. Hopefully Nikon will implement DGO in the next versions of their cameras.

Note that Panasonic implements DGO for its video as well, which unlike the stills implementation can be controlled with a user setting they call "DR Boost".
 
Last edited:
Interesting - thanks for the testing and your report!
 
I have read the internet "hysteria" :-(

As an actual owner of a Z6 III the reduction is slight - at low ISOs less than 10% of the dynamic range of of a Z6 II - and still over 25% more than the dynamic range of an everyday sunlit scene :-D;-)

At very high ISO's most cameras dynamic range often falls below that of an everyday sunlit scene.
 
Will Nikon use that Panasonic sensor for a "Z6iii Mark 2" thereby eliminating the dynamic range problem at low ISOs?
 
I have read the internet "hysteria" :-(

As an actual owner of a Z6 III the reduction is slight - at low ISOs less than 10% of the dynamic range of of a Z6 II - and still over 25% more than the dynamic range of an everyday sunlit scene :-D;-)

At very high ISO's most cameras dynamic range often falls below that of an everyday sunlit scene.
Agreed, it's just hype:

Animation: Panasonic S1 II vs Z6 III, ISO 100 -7EV
 
still over 25% more than the dynamic range of an everyday sunlit scene
I have seen some significantly divergent opinions on this forum regarding how many stops there are in "an everyday sunlit scene" (some of those opinions make mention of "black velvet"), and there's the rub. So personally I am skeptical of any claimed percentages (such as "25% more") based upon such an ill-defined denominator.

Thanks Horshack for this fascinating writeup. I have no doubt that what you have said is meticulously researched, and follows the facts wherever they may lead, even if that makes some members of this forum unhappy.

DGO appears to be very promising. Here's hoping Nikon explores this promising technology further, and manages to eke more DR out of their partially stacked sensor, now that Panasonic has done so.

--
Jonathan
https://www.flickr.com/photos/jtr27/
 
Last edited:
Will Nikon use that Panasonic sensor for a "Z6iii Mark 2" thereby eliminating the dynamic range problem at low ISOs?
I theorized in my S1 II analysis thread they're using the same version sensor but that Nikon's ISP (ie, EXPEED 7) just doesn't support Sony's "Clear HDR" sensor feature that Panasonic is likely using for the DGO implementation.
 
Last edited:
Is the ISO 800 amplification done at the pixel, before the sensor scan sends it to the camera processor? Is there just one scan of the sensor, and this dual ISO processing happens later?
 
...a few weeks ago I acquired an S1 II and started my own deep technical analysis. That analysis revealed evidence Panasonic was actually employing a dual gain output to achieve the S1 II's excellent DR. This is where the camera performs two readouts of a single exposure, one at ISO 100 and the other ISO 800, then merges the two in the resulting raw. This combines the low ISO saturation capacity of ISO 100 with the deep shadow noise improvement seen at the dual gain step of ISO 800.
So, help me understand the mechanics, if you will, of how something like this would work. What I'm envisioning is an exposure made at some shutter speed, X, at ISO 100. Part way through the exposure at 1/8*X into the shutter actuation, the camera buffers an image at the dual-gain ISO to have that lower level of read noise temporarily stored. When the full exposure time, X, has passed and the ISO 100 image is processed, it's merged with the buffered ISO 800 image to create a composite merging the shot noise of the longer exposure and the read noise of the shooter exposure.

Is that at all approaching the reality of how such tech would work? Or is it more like a dark frame in astrophotography; a frame of nothing that maps read noise at a given ISO so the noise can be subtracted from an exposure of a celestial object?

--
Bill Ferris Photography
https://billferrisphotography.pixieset.com/arizonaslittleserengeti/
 
Last edited:
This technology was first introduce by Lumix in either the M43's GH6 or G9ii. I know the GH6 had a few issues regarding image quality, however this was address with the G9ii and later added to the GH7.

To be honest I didn't notice much difference between my G9 and G9ii. However, that's probably a good thing when considering the G9ii had an extra 5-megapixels. This could be useful tool when companies could use the same approach for 80 or even 100-megapixel files or even global shutter in the future.
 
...a few weeks ago I acquired an S1 II and started my own deep technical analysis. That analysis revealed evidence Panasonic was actually employing a dual gain output to achieve the S1 II's excellent DR. This is where the camera performs two readouts of a single exposure, one at ISO 100 and the other ISO 800, then merges the two in the resulting raw. This combines the low ISO saturation capacity of ISO 100 with the deep shadow noise improvement seen at the dual gain step of ISO 800.
So, help me understand the mechanics, if you will, of how something like this would work. What I'm envisioning is an exposure made at some shutter speed, X, at ISO 100. Part way through the exposure at 1/8*X into the shutter actuation, the camera buffers an image at the dual-gain ISO to have that lower level of read noise temporarily stored.
Why would you shorten the exposure by three stops when this part of the process serves to improve the shadows?
When the full exposure time, X, has passed and the ISO 100 image is processed, it's merged with the buffered ISO 800 image to create a composite merging the shot noise of the longer exposure and the read noise of the shooter exposure.
 
I have seen some significantly divergent opinions on this forum regarding how many stops there are in "an everyday sunlit scene" (some of those opinions make mention of "black velvet"),
Correct - Kodak published a range of subject reflectances of subjects commonly photographed in sunlight - almost 50 years ago - in 1947.

Since then nobody seems to have seriously challenged the scientifically based 8 step scene range :-)

Black velvet
is the darkest subject used by Kodak in which detail could be recorded (percentage reflection 0.6%) as distinct from the blackest objects in shadow where detail is not recorded using the Kodak scale.

Black velvet is slightly darker than blackest objects in sunlight - percentage reflectance 1.3%).

Nearer the middle of the reflectance range are subjects such as light grass and foliage with 20% reflectance and medium/dark grass with around 10% reflectance.

The Kodak Neutral Test Card (gray side) has 18% reflection.

The whitest subjects in which detail are recorded using the Kodak system are the whitest clouds, whitest snow and bright white paint (percentage reflection 80%) - as distinct from light sources and specular (sunlight) reflections.

All this covers an 8 stop exposure range.

Being able to detect at the same time detail in both the whitest clouds and black velvet in sunlight is challenging for human vision.

At low ISO's the Z6 III has several stops more than 8 DR - the debate seems manly about a total of 11.5 v 12.0 stops - depending on the sensor measuring methodology.

At base ISO the Z6 III has 3-4 stops DR in reserve (compared to the "Kodak" 8 stops) to help record the blackest objects in shadow or light sources - when required.

Back to black velvet - it is a recognisable subject reflecting a little less than 1% of the sunlight falling on a scene. Can you think of an alternative subject with less than 1% reflectance ?

--
Leonard Shepherd
In lots of ways good photography is similar to learning to play a piano - it takes practice to develop skill in either activity.
 
Last edited:
"Eight stops of DR is all you need" makes me chuckle. Why? Because it reminds me a bit of "four megapixels of resolution is all you need," a claim vehemently repeated and aggressively defended by enthusiasts, not all that long ago. Enthusiasts who, by uncanny coincidence, happened to own four megapixel cameras.

Those same enthusiasts no doubt also dismissed suggestions that six megapixels might be a useful improvement, as "internet hysteria."

Of course, people no longer say "four megapixels is all you need," because the people who said it have long ago quietly purchased 24 megapixel cameras, or 36, or 45, and are busy using them.

I was disappointed but not surprised to see a word like "hysteria" appear in this thread. I did not think it was helpful or open-minded. It struck me as a defensive response, a transparent attempt to sweep some "inconvenient" findings under the rug.

Back to DR. Of course it is of interest to the community when a new camera comes out with less DR than its predecessor. And of course it is also of interest when another company gets more DR out of the same sensor. And whatever Kodak said in 1923, DR exceeding eight stops is beneficial to modern digital photographers. Of course it is.

"In lots of ways good photography is similar to learning to play a piano - it takes practice to develop skill in either activity." It also takes curiosity and open mindedness. We should be able to discuss findings like Horshack's as grownups, without reflexively disparaging them.

--
Jonathan
https://www.flickr.com/photos/jtr27/
 
Last edited:
"Eight stops of DR is all you need" makes me chuckle.
Why?

8 stops is but one standard - though widely encountered in everyday sunlit photography.

8 stops is enough for a typical scene in sunlight - except for the blackest objects in shadow.

The greater 11.5 or 12.5 stops DR that many recent digital cameras achieve shooting RAW (less with jpg) enables a greater digital image dynamic range - though highlight detail remains easily blown out.

Testing methodology varies - with different scores as a result.

To get more than 15 stops DR in a digital image generally requires a medium format size sensor.

I am not aware of any printing paper that can accommodate 10 stops DR without some tone compression - though this is another topic.

--
Leonard Shepherd
In lots of ways good photography is similar to learning to play a piano - it takes practice to develop skill in either activity.
 
Last edited:
"Eight stops of DR is all you need" makes me chuckle.
Why?

8 stops is but one standard - though widely encountered in everyday sunlit photography.

8 stops is enough for a typical scene in sunlight - except for the blackest objects in shadow.

The greater 11.5 or 12.5 stops DR that many recent digital cameras achieve shooting RAW (less with jpg) enables a greater digital image dynamic range - though highlight detail remains easily blown out.

Testing methodology varies - with different scores as a result.

To get more than 15 stops DR in a digital image generally requires a medium format size sensor.

I am not aware of any printing paper that can accommodate 10 stops DR without some tone compression - though this is another topic.
 
"Eight stops of DR is all you need" makes me chuckle.
Why?

8 stops is but one standard - though widely encountered in everyday sunlit photography.
That seems a bit low. as the dynamic range of an average daylight scene can be estimated as 18 stops when referencing PanoTools.org. Other sources show it as a bit lower, but I think it might be more accurate to say the dynamic range of an average daylight scene is somewhere around 12-18 stops. Lots of variables at play. For example, a scene at the beach with no major shadows or reflective surfaces might be close to 8 stops at high noon.

If your camera sensor is actually limited to 8 stops of DR, and you shoot lots of different daylight scenes, you're going to miss out on capturing many a detail in your shots
 
...a few weeks ago I acquired an S1 II and started my own deep technical analysis. That analysis revealed evidence Panasonic was actually employing a dual gain output to achieve the S1 II's excellent DR. This is where the camera performs two readouts of a single exposure, one at ISO 100 and the other ISO 800, then merges the two in the resulting raw. This combines the low ISO saturation capacity of ISO 100 with the deep shadow noise improvement seen at the dual gain step of ISO 800.
So, help me understand the mechanics, if you will, of how something like this would work. What I'm envisioning is an exposure made at some shutter speed, X, at ISO 100. Part way through the exposure at 1/8*X into the shutter actuation, the camera buffers an image at the dual-gain ISO to have that lower level of read noise temporarily stored. When the full exposure time, X, has passed and the ISO 100 image is processed, it's merged with the buffered ISO 800 image to create a composite merging the shot noise of the longer exposure and the read noise of the shooter exposure.

Is that at all approaching the reality of how such tech would work? Or is it more like a dark frame in astrophotography; a frame of nothing that maps read noise at a given ISO so the noise can be subtracted from an exposure of a celestial object?
It's a great question and one that I spent some time exploring in my S1 II analysis thread. Of the two methods you described, one approaching the first would be more applicable. I haven't found any patents from Sony on their technique but I did find one from Samsung that is along the lines of what you suggested. I posted about it here and here.

One interesting clue for at least the sensor readout method is the fact that Panasonic's implementation increases the readout time by approximately 2.3x times. This is based on myhigh-frequency LED rolling shutter measurement methodology. This is for the S1 II's video mode (electronic shutter), which by all indications use an identical technique to its stills mode. I had to measure this in video mode since the mechanical shutter-only implementation for stills hides the readout phase from analysis. I've theorized there's no technical limitation preventing Panasonic from supporting DGO for stills with the S1 II's electronic shutter - it seems more a decision based on the impact to readout time, maybe pushing rolling shutter beyond the threshold they consider acceptable for stills photography.

The fact we can measure a 2.3x increase in readout time indicates that at least the readout of both the low and high gain are occurring in parallel (or at least interleaved). This is because the readout time is measured by counting the number of on/off cycles from my LED light source, so if the readout were done by reading out the entire sensor in low gain and then high gain then the visual presentation of the light bands would be overlapped and interleaved due to the phase shift of cycles between the two readouts.
 
This technology was first introduce by Lumix in either the M43's GH6 or G9ii. I know the GH6 had a few issues regarding image quality, however this was address with the G9ii and later added to the GH7.

To be honest I didn't notice much difference between my G9 and G9ii. However, that's probably a good thing when considering the G9ii had an extra 5-megapixels. This could be useful tool when companies could use the same approach for 80 or even 100-megapixel files or even global shutter in the future.
Yep, I bought a used GH7 specifically to compare against the S1 II's method. The problem I'm having in that effort is unlike the G9 II, the GH7 appears to support its DGO for all shutter speeds, so I don't have a non-DGO baseline on the same sensor to compare it to. I would use video for that but bth the G9 II and GH7 use DGO for all modes as well, so again there's no non-DGO measurement possible on the same sensor for me to compare it to..
 
Last edited:
I have read the internet "hysteria" :-(

As an actual owner of a Z6 III the reduction is slight - at low ISOs less than 10% of the dynamic range of of a Z6 II - and still over 25% more than the dynamic range of an everyday sunlit scene :-D;-)

At very high ISO's most cameras dynamic range often falls below that of an everyday sunlit scene.
Agreed, it's just hype:

Animation: Panasonic S1 II vs Z6 III, ISO 100 -7EV
ISO 100 is not meant to be used like this, underexposed 7 stops. I am not even sure if this experiment has been done correctly, because ISO100 is not the same on all cameras. Could be that the Z6III at ISO100 can be exposed brighter without blowing the highlights in the same scene. There is at most a 1-stop difference in dynamic range up to ISO400.

Here is what ISO12800 looks like on these cameras. I think that is a more realistic experiment.

2518fb146e9a4eaa85cc9f2c058f2899.jpg.png
 
I have seen some significantly divergent opinions on this forum regarding how many stops there are in "an everyday sunlit scene" (some of those opinions make mention of "black velvet"),
Correct - Kodak published a range of subject reflectances of subjects commonly photographed in sunlight - almost 50 years ago - in 1947.

Since then nobody seems to have seriously challenged the scientifically based 8 step scene range :-)

Black velvet
is the darkest subject used by Kodak in which detail could be recorded (percentage reflection 0.6%) as distinct from the blackest objects in shadow where detail is not recorded using the Kodak scale.

Black velvet is slightly darker than blackest objects in sunlight - percentage reflectance 1.3%).

Nearer the middle of the reflectance range are subjects such as light grass and foliage with 20% reflectance and medium/dark grass with around 10% reflectance.

The Kodak Neutral Test Card (gray side) has 18% reflection.

The whitest subjects in which detail are recorded using the Kodak system are the whitest clouds, whitest snow and bright white paint (percentage reflection 80%) - as distinct from light sources and specular (sunlight) reflections.

All this covers an 8 stop exposure range.

Being able to detect at the same time detail in both the whitest clouds and black velvet in sunlight is challenging for human vision.

At low ISO's the Z6 III has several stops more than 8 DR - the debate seems manly about a total of 11.5 v 12.0 stops - depending on the sensor measuring methodology.

At base ISO the Z6 III has 3-4 stops DR in reserve (compared to the "Kodak" 8 stops) to help record the blackest objects in shadow or light sources - when required.

Back to black velvet - it is a recognisable subject reflecting a little less than 1% of the sunlight falling on a scene. Can you think of an alternative subject with less than 1% reflectance ?
So your claim cameras only need 8EV of dynamic range is based on an imaginary existence where only diffusely-reflective subjects exist? Is that some type of post-nuclear apocalypse world where direct sunlight no longer occurs due to particulates from a nuclear winter? And where all power-generation sources have been destroyed so direct-emitting light sources no longer work?

There's a name for that scenario - it's called SDR, Standard Dynamic Range. It was the standard for imaging for many years, not because it's all that's required but because it was all that was technically possible. Imaging has since moved on to HDR, which explicitly adds specular highlights and direct-emission into the range of luminance targeted.

But if you're only interested in SDR scenarios then by all means the Z6 III is perfectly suitable. In fact, nearly every camera released since the dawn of the digital age would suit that role admirably.
 

Keyboard shortcuts

Back
Top