j4nu wrote:
In my understanding, it's more or less the same/similar idea, but the implementation is different. DGO is a low-level technique I'd say (you get the end result in a ~single exposure time), while what phones do is merging full exposures (i.e. you need to wait longer, so that multiple exposures are done).
Employing multiple exposures only works well when there is very little movement. DGO only captures pixel data once, so it is free of time blending artifacts.
Daran wrote:
Employing multiple exposures only works well when there is very little movement. DGO only captures pixel data once, so it is free of time blending artifacts.
My thinking as well. It's probably unrealistic to think that computational imagery from smartphones would in some future come to real cameras - at least in the same way they work now. Imagine A9III blending multiple exposures together while shooting 30 or 120 frames per second and tracking subject.. It's more probable that similar approaches need to be develop from a different point of view and DGO might been one step into that direction. Anyway, from what I understand the DGO seems like a genuine way to increase dynamic range.
A7V is a confusing camera.
First, if the dual output gain obscures the effect of baked-in noise reduction, then Bill Claff's replacing the inverted triangles by circles is questionable. The 1.5 DR stops difference between the mechanical and electronic shutters is surprisingly large. I am not sure this can be explained by the dual output gain alone. Basically, now it is impossible to tell whether A7V employs some form of noise reduction, or does something else entirely.
Other things invite questions. Like, why the dual output gain couldn't be implemented with the ES? This is not a technical impossibility in general, as demonstrated by Panasonic G9II. Next, why the "conventional" dual-gain ISO changed from 400 on A7IV to 1000 on A7V? Also, why this has no effect on the DR at ISO 1000 with the MS? Should this be understood as implying that the dual output gain is active only in the 100 - 800 ISO range? Is this just a coincidence that the two DR curves, with the ES and MS shutters, merge at 1000 ISO?
The last thing is that in science important findings are not widely accepted until independently confirmed. The work of Bill Claff is highly respected, yet he made an unexplained mistake when he reported the PDR of Panasonic G9II with ES. I am willing to trust Bill, yet it is important to collect more observations on the DR of A7V. As I already mentioned this, if I had the camera, I would've collected a series of dark frames with the ES and MS, at base ISO, while warying shutter speeds in the 1s to 1/250s range for the ES. Some real-life examples of shadow recovery with ES vs MS should also be useful,again with relatively fast SS, e.g. from 1/60s.
ruthenium wrote:
A7V is a confusing camera.
First, if the dual output gain obscures the effect of baked-in noise reduction, then Bill Claff's replacing the inverted triangles by circles is questionable. The 1.5 DR stops difference between the mechanical and electronic shutters is surprisingly large. I am not sure this can be explained by the dual output gain alone.
THe S1 II's DGO has 1.3EV more DR than the non-DGO Z6 III even though they share the same Sony 24MP partially-stacked sensor:
A valid point.
Also of interest is that the PDR of A7V with mechanical shutter at 400 ISO is practically the same as the PDR of A7IV at this ISO, presumably with MS as well. Thus, I am not sure when and in what ISO range the DGO of A7V operates.
ruthenium wrote:
A valid point.
Also of interest is that the PDR of A7V with mechanical shutter at 400 ISO is practically the same as the PDR of A7IV at this ISO, presumably with MS as well. Thus, I am not sure when and in what ISO range the DGO of A7V operates.
Edited my reply: Presumably it operates up until the dual gain ISO, since that's the point the tapering benefit would completely disappear. The only way to tell for sure would be to measure the readout time of each ISO to see where the 2x readout ends, but measuring that obviously is nearly impossible with the mechanical shutter.
ruthenium wrote:
Should this be understood as implying that the dual output gain is active only in the 100 - 800 ISO range? Is this just a coincidence that the two DR curves, with the ES and MS shutters, merge at 1000 ISO?
If the first amplifier is based on ISO 100 and second one 800, then yes, the DGO will only be meaningful between those. Think it as an interpolation. Beyond 800, even if DGO is used, the weight on the first will be 0 and the weight on the second will be 1.
Edit: after some second thought, I'm not sure simple interpolation is what's happening at low ISO.
I also do wonder why ES can't use this. Is it because there is no shutter screen during ES, so that the values read during the second read will be even higher? Why can't they read once and send to both amplifier at the same time? Maybe they have some data path limitations that they can resolve in future.
snapsy wrote:
Edited my reply: Presumably it operates up until the dual gain ISO, since that's the point the tapering benefit would completely disappear. The only way to tell for sure would be to measure the readout time of each ISO to see where the 2x readout ends, but measuring that obviously is nearly impossible with the mechanical shutter.
I don't think the 2x read speed during ES can be measured? If the pixel values stayed the same between the two reads, then the rolling shutter effect to derive the read speed will only reflect 1x read speed. if the pixel values continue to accumulate between the two reads, then DGO is only accurate for long shutter speeds where the delay between the two reads is negligible compared to the shutter speed.
tctmp wrote:
I don't think the 2x read speed during ES can be measured? If the pixel values stayed the same between the two reads, then the rolling shutter effect to derive the read speed will only reflect 1x read speed. if the pixel values continue to accumulate between the two reads, then DGO is only accurate for long shutter speeds where the delay between the two reads is negligible compared to the shutter speed.
The 2x read speed is based on video mode measurements, where on the S1 II's DGO can be turned on/off with a user setting.
snapsy wrote:
The 2x read speed is based on video mode measurements, where on the S1 II's DGO can be turned on/off with a user setting.
But what's the observation? You have double number of the bands when it's on? if that's the case, does that mean the two reads are not one read across the sensor, then another read across the sensor. Rather, one read across say 8 rows, then another read across the same 8 rows, then move on to the next 8 rows? I suppose that makes some sense. In that case, I may classify it as a single read that just take longer to process since it needs to drive both amplifiers. But in that case, I can't see what's stopping it from being enabled in ES.
It will be a less confusing camera when people start using it to make pictures, rather than treating it like a science project.
ruthenium wrote:
A7V is a confusing camera.
First, if the dual output gain obscures the effect of baked-in noise reduction, then Bill Claff's replacing the inverted triangles by circles is questionable. The 1.5 DR stops difference between the mechanical and electronic shutters is surprisingly large. I am not sure this can be explained by the dual output gain alone. Basically, now it is impossible to tell whether A7V employs some form of noise reduction, or does something else entirely.
Other things invite questions. Like, why the dual output gain couldn't be implemented with the ES? This is not a technical impossibility in general, as demonstrated by Panasonic G9II. Next, why the "conventional" dual-gain ISO changed from 400 on A7IV to 1000 on A7V? Also, why this has no effect on the DR at ISO 1000 with the MS? Should this be understood as implying that the dual output gain is active only in the 100 - 800 ISO range? Is this just a coincidence that the two DR curves, with the ES and MS shutters, merge at 1000 ISO?
The last thing is that in science important findings are not widely accepted until independently confirmed. The work of Bill Claff is highly respected, yet he made an unexplained mistake when he reported the PDR of Panasonic G9II with ES. I am willing to trust Bill, yet it is important to collect more observations on the DR of A7V. As I already mentioned this, if I had the camera, I would've collected a series of dark frames with the ES and MS, at base ISO, while warying shutter speeds in the 1s to 1/250s range for the ES. Some real-life examples of shadow recovery with ES vs MS should also be useful,again with relatively fast SS, e.g. from 1/60s....Show more →
tctmp wrote:
But what's the observation? You have double number of the bands when it's on? if that's the case, does that mean the two reads are not one read across the sensor, then another read across the sensor. Rather, one read across say 8 rows, then another read across the same 8 rows, then move on to the next 8 rows? I suppose that makes some sense.
Double the bands. I've done a bunch of experiments on the S1 II to answer the very question you're asking, which included a very high-frequency banding generator (running on code I wrote on an Arduino) to check for any stratification of reads between rows/groups of rows.
snapsy wrote:
Double the bands. I've done a bunch of experiments on the S1 II to answer the very question you're asking, which included a very high-frequency banding generator (running on code I wrote on an Arduino) to check for any stratification of reads between rows/groups of rows.
But then I don't see why ES can't be enabled. Other than 2x the rolling shutter, I don't see other ill effects or extra work needed. Just let users choose then.
