p.3 #1 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Mike Jacks0n wrote:
Are you saying you don't think the R5 benefits from 14 bit capture in the 200-400 range or are you saying the A9 II doesn't benefit from it and that range?
Also, any NR issues you are worried about on the R5 I don't really notice. That's not to say it doesn't exist, just to say its very well controlled and a lack of detail is not an issue that I'm finding.
It is very well controlled. This NR is only there to artificially boost DR. If it isn't measured one could try to check stars in the sky.
Sony has a star eater problem when longer exposures are taken (one can see some stars vanishing), but Sony doesn't use NR at exposures < 4s, I think. A comparison to the R5 at pixel level could possibly reveal filtered stars?
p.3 #2 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Btw: from what I read there is no evidence that R5 performs NR to its RAWs. Bill from the dpreview forum noticed something unusual in the raws and believes might be NR. Other highly regarded people like Illiah Borg (author ob lib raw) is rather reluctant and doesn't call it NR, since it might as well be something else. Hence I find it rather hasty to conclude the R5 'cheats' ...
p.3 #3 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
osv2 wrote:
i haven't seen any links proving that 14 bit is better at iso200-iso400, can you re-post whatever it is that you are referring to.
The dpreview studio comparison photo's visually show electronic is worse than mechanical in dynamic range. Electronic is 12 bit. Mechanical is 14 bit. I am assuming that the bit rate is causing it.
"
Interesting to look at the review of R6 - it shows that using electronic shutter (12bit) drops the ability to push shadows by about 1- 2 stops (my assessment) vs mechanical shutter (14 bit) non plus (8fps).
This lessor performance of a9 at iso100-400 is also shown in comparisons between e shutter and mechanical shutter for a9 vs R5. I am assuming a9 is at 12 bit continuous in both modes.
p.3 #4 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
phibes wrote:
Btw: from what I read there is no evidence that R5 performs NR to its RAWs. Bill from the dpreview forum noticed something unusual in the raws and believes might be NR. Other highly regarded people like Illiah Borg (author ob lib raw) is rather reluctant and doesn't call it NR, since it might as well be something else. Hence I find it rather hasty to conclude the R5 'cheats' ...
Bill's FFTs show correlation between pixels, which by definition is noise reduction. Whether or not Canon's goal in applying the kernel that produced that correlation was NR is what's up for debate. I believe they're applying it to hide an image artifact such as banding - this is based on how the amount of measured NR decreases as the noise level increases (increasing ISO or base-ISO long exposures).
I posted frequency plots of R5 blackframes here, including a comparison to Z6 blackframes, which don't exhibit NR. I posted Canon M100 plots here.
Upon further investigation I discovered that the masked border pixels around the R5's active area do not exhibit the NR, which provided a means to establish the true noise levels off the sensor. Here is a plot comparing the noise stdev between the active and border areas of the R5 sensor at each ISO: https://photos.smugmug.com/photos/i-gT4Mvx3/0/d3e806e0/O/i-gT4Mvx3.png
I then postulated that the border pixels could be use to reverse-engineer the imaging kernel being applied for convolution, since 5 of the 9 "true" pixel values would be available at each corner of the active imaging area (5 border pixels and 4 active pixels). A poster on dpreview had some success in reverse engineering the kernel here.
p.3 #5 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
snapsy wrote: Bill's FFTs show correlation between pixels, which by definition is noise reduction. Whether or not Canon's goal in applying the kernel that produced that correlation was NR is what's up for debate. I believe they're applying it to hide an image artifact such as banding - this is based on how the amount of measured NR decreases as the noise level increases (increasing ISO or base-ISO long exposures).
I posted frequency plots of R5 blackframes here, including a comparison to Z6 blackframes, which don't exhibit NR. I posted Canon M100 plots here.
Upon further investigation I discovered that the masked border pixels around the R5's active area do not exhibit the NR, which provided a means to establish the true noise levels off the sensor. Here is a plot comparing the noise stdev between the active and border areas of the R5 sensor at each ISO: https://photos.smugmug.com/photos/i-gT4Mvx3/0/d3e806e0/O/i-gT4Mvx3.png
I then postulated that the border pixels could be use to reverse-engineer the imaging kernel being applied for convolution, since 5 of the 9 "true" pixel values would be available at each corner of the active imaging area (5 border pixels and 4 active pixels). A poster on dpreview had some success in reverse engineering the kernel here....Show more →
Nice link and read!
Sep 17, 2020 at 03:41 PM
osv2 Offline • • • • [X]
p.3 #6 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Scott Stoness wrote:
The dpreview studio comparison photo's visually show electronic is worse than mechanical in dynamic range. Electronic is 12 bit. Mechanical is 14 bit. I am assuming that the bit rate is causing it.
"
Interesting to look at the review of R6 - it shows that using electronic shutter (12bit) drops the ability to push shadows by about 1- 2 stops (my assessment) vs mechanical shutter (14 bit) non plus (8fps).
This lessor performance of a9 at iso100-400 is also shown in comparisons between e shutter and mechanical shutter for a9 vs R5. I am assuming a9 is at 12 bit continuous in both modes....Show more →
EDIT: i tried to substantiate it either way with a9ii electronic shutter data, but a9ii es apparently wasn't tested on the bill claff pdr site.
from that dpr r6 review link: "The other factor worth being aware of is that the detail appears to smear a little if you push all the way down into the deep shadows. It appears Canon is applying some noise reduction to the darkest regions of its Raw files, **resulting in a loss of detail**. Whether this is preferable to losing those tones to noise is a matter of taste, but we're always disappointed to see noise reduction applied to Raw data without an option to turn it off."
r5 will also show loss of detail, because the bill claff r5 pdr data shows noise reduction being applied all the way up through iso636 at least.
you say that you can't see the loss of detail, but dpr says that it's visible with the r6, it's probably going to be visible with the r5 as well, at the lowest dark levels.
"...dynamic range... So there's only a slight difference if you compare the ISO 400 files but if you push further down into the shadows by looking at the ISO 100 and 200 images, the difference becomes very apparent." https://www.dpreview.com/reviews/canon-eos-r6-review/6
you want to shoot at iso200-iso400, where 14 bit isn't going to make much if any d.r. difference with the r6, and most likely not with the r5 either.
p.3 #7 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
BirdShooter7 (Greg Lavaty) shared his hummingbird image. https://www.dpreview.com/forums/post/64374150
It shows that the Canon R6 reads out rows in groups of 8 just like the 1DX3. This is as expected since they seem to share the sensor.
p.3 #8 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
snapsy wrote: Bill's FFTs show correlation between pixels, which by definition is noise reduction. Whether or not Canon's goal in applying the kernel that produced that correlation was NR is what's up for debate. I believe they're applying it to hide an image artifact such as banding - this is based on how the amount of measured NR decreases as the noise level increases (increasing ISO or base-ISO long exposures).
I posted frequency plots of R5 blackframes here, including a comparison to Z6 blackframes, which don't exhibit NR. I posted Canon M100 plots here.
Upon further investigation I discovered that the masked border pixels around the R5's active area do not exhibit the NR, which provided a means to establish the true noise levels off the sensor. Here is a plot comparing the noise stdev between the active and border areas of the R5 sensor at each ISO: https://photos.smugmug.com/photos/i-gT4Mvx3/0/d3e806e0/O/i-gT4Mvx3.png
I then postulated that the border pixels could be use to reverse-engineer the imaging kernel being applied for convolution, since 5 of the 9 "true" pixel values would be available at each corner of the active imaging area (5 border pixels and 4 active pixels). A poster on dpreview had some success in reverse engineering the kernel here....Show more →
Very interesting data. If camera manufacturers don't want people like you figuring out what they are doing they need to do a better job hiding stuff.
The data on Claff's site for the R5 indicates that the noise reduction is applied between ISO 100 and 640 (down turned triangles). Three data points on your chart are for ISO settings that have noise reduction (100, 200, 400). The data for ISO 800 and above do not have noise reduction, according to Claff, but you still see noise reduction for the active pixels compared to masked pixels. Why is that? Are you of the opinion that the noise reduction is still being applied above ISO 800?
p.3 #9 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
dclark wrote:
Very interesting data. If camera manufacturers don't want people like you figuring out what they are doing they need to do a better job hiding stuff.
The data on Claff's site for the R5 indicates that the noise reduction is applied between ISO 100 and 640 (down turned triangles). Three data points on your chart are for ISO settings that have noise reduction (100, 200, 400). The data for ISO 800 and above do not have noise reduction, according to Claff, but you still see noise reduction for the active pixels compared to masked pixels. Why is that? Are you of the opinion that the noise reduction is still being applied above ISO 800? ...Show more →
The FFT frequency graphs are a bit more sensitive to pixel correlations than the FFT power spectrum plots that Bill is using, the later of which require careful visual interpretation as the strength of the correlations drop. The graphs show NR up through around ISO 1600 but at that point they're barely discernible and not statistically significant, so Bill's cutoff near ISO 800 are about right for practical use.
The stdev of the border pixels still diverge vs active pixels at around 10% through ISO 25,600. I'm assuming that's attributable to some baseline noise differences related to the structure of the border/masked pixels but I haven't spent much time looking into that.
p.3 #10 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Stan (aka stanj) provided an R5 DNG image of a hummer taken with ES that shows the bands and streaks. I measured them to be 8 rows, just as for the 1DX3 and R6. Thanks Stan.
So it seems that all the Canons that have fast ES are reading out rows in groups of 8, and all seem to have shutter readout in the range of 1/50 to 1/60 sec, although I have yet to see an accurate measurement of the shutter readout speed for any of them.
p.3 #11 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Sometimes I read technical analysis of a camera and I chuckle at the conclusions. Phrases like "unusable" and "major distortions" get tossed around. Early on, the consensus with the R5 from "armchair engineers" was that the electronic shutter performance was "terrible"...I actually saw multiple people saying the performance was terrible and unusable. Granted, I've spoken with several NFL photographers that have shown me examples and told me that the performance for their use case would indeed be just that - unusable. But that doesn't mean it won't work for other use cases...please see bellow.
What do I deem "unusable"? When the rolling shutter effect takes away from the subject of the image. In the case of the R5 as a motorsports photographer, I find the electronic shutter to be HIGHLY usable and extremely beneficial.
Two weeks ago, I had the opportunity to use my new R5 to cover one of our races that was thankfully able to still be held. I used electronic shutter extensively as I threw everything I could at the camera...I left the event prepared to sell my 1DX Mark II...and I have. Here are just a few examples taken with the R5 in the 12-bit RAW electronic shutter that some people are afraid to use. Don't be afraid. Give it a shot and see if you think it will work for the application you're working in. LEARN how your camera performs and understand when you can exploit this feature - I think you'll be very surprised.
Distortion is there, but I don't consider this shot ruined because I used it and most people wouldn't notice the distortion if it wasn't pointed out to them or they were looking for it. MS-NMRA_NMCA-MARTIN2020-0229 by LSXPhotog, on Flickr
This is a single shot from a 65 image sequence I shot using the Electronic Shutter in the R5. Distortion is visible, but it doesn't really take away from this sequence of images. The main point is that I was able to capture a brilliant series of images with the car tracked in every frame. We're using the sequence to create a two-page spread that should look really awesome. Using 20fps now lets us choose WHICH images we want to use to show what happened. It's awesome. MS-NMRA_NMCA-MARTIN2020-0653 by LSXPhotog, on Flickr
p.3 #12 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
I would also like to point out that using the 12-bit electronic shutter under these particular artificial lights didn't create any banding from the light frequency. High ISO performance also remains extremely impressive! This is a 45mp image sensor using electoronic shutter under artificial light at ISO 12,800. Not only is is completely usable and I delivered to the client, but it looks downright exceptional. So my apologies if I'm laughing at people who say you can't use the camera in the 12-bit electronic shutter...that sentiment is incorrect.
I'm a Canon user, and acknowledge my bias...but this has to be the best performing photography camera on the market with a truly incredible sensor.
p.3 #13 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Visceral Image wrote:
Sometimes I read technical analysis of a camera and I chuckle at the conclusions. Phrases like "unusable" and "major distortions" get tossed around. Early on, the early consensus with the R5 from "armchair engineers" was that the electronic shutter performance was "terrible"...I actually saw multiple people saying the performance was terrible and unusable. Granted, I've spoken with several NFL photographers that have shown me examples and told me that the performance for their use case would indeed be just that - unusable. But that doesn't mean it won't work for other use cases...please see bellow.
In other words, there are situations where the R5's electronic shutter is perfectly useable, and some where it's not. I'm not seeing the contradiction.
p.3 #14 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Visceral Image wrote:
I would also like to point out that using the 12-bit electronic shutter under these particular artificial lights didn't create any banding from the light frequency. High ISO performance also remains extremely impressive! This is a 45mp image sensor using electoronic shutter under artificial light at ISO 12,800. Not only is is completely usable and I delivered to the client, but it looks downright exceptional. So my apologies if I'm laughing at people who say you can't use the camera in the 12-bit electronic shutter...that sentiment is incorrect.
I'm a Canon user, and acknowledge my bias...but this has to be the best performing photography camera on the market with a truly incredible sensor.
As Snapsy said, sometimes it works, sometimes, not. As you are outside at iso12800 the artificial light (whose frequency we don't know and which is important) wasn't a major light source, too.
In many venues and churches a readout speed of 1/60s isn't adequate, in my experience as banding is possible. Outside it may be fully sufficient to use the E-shutter. As an event shooter I know what my A9ii is capable of and although its read out speed in E-shutter is unmatched, I always need to be careful and check beforehand.
p.3 #15 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
How is it a weird post? If the lights were not on, the image would be completely black...so the artificial lights were the only source of light. I provided an example where it worked just fine.
Holger wrote:
Weird post.
As Snapsy said, sometimes it works, sometimes, not. As you are outside at iso12800 the artificial light (whose frequency we don't know and which is important) wasn't a major light source, too.
In many venues and churches a readout speed of 1/60s isn't adequate, in my experience as banding is possible. Outside it may be fully sufficient to use the E-shutter. As an event shooter I know what my A9ii is capable of and although its read out speed in E-shutter is unmatched, I always need to be careful and check beforehand.
p.3 #16 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Exactly. With drag racing, getting cars off the line isn't going to be a situation where ES is going to show artifacts. Neither will the cars if you are tracking them before the 1/8 marker panning with them. However panning a vehicle once it has entered triple digit speeds will likely show artifacts in the background, all those nice straight vertical lines may find themselves leaning over a bit.
Sep 18, 2020 at 10:42 AM
osv2 Offline • • • • [X]
p.3 #17 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Visceral Image wrote:
Sometimes I read technical analysis of a camera and I chuckle at the conclusions. Phrases like "unusable" and "major distortions" get tossed around. Early on, the consensus with the R5 from "armchair engineers" was that the electronic shutter performance was "terrible"...I actually saw multiple people saying the performance was terrible and unusable. Granted, I've spoken with several NFL photographers that have shown me examples and told me that the performance for their use case would indeed be just that - unusable. But that doesn't mean it won't work for other use cases.
it's good that you sort of understand the manner in which your r5 is crippled, so that you don't attempt to use it where it doesn't work.
see the r5 distorted football shot below...
Visceral Image wrote:
This is a single shot from a 65 image sequence I shot using the Electronic Shutter in the R5. Distortion is visible, but it doesn't really take away from this sequence of images.
i agree with that, it's not much of a factor with thin-dof panning, plus you had a target-rich environment, so a couple of those pics are winners, thx for posting 'em... i've shot thousands of drag racing pics, so i appreciate seeing it.
i think that the head-on shot is a fail due to the lack of dof tho, it just doesn't work when you need to see tire wrinkle and such... you used panning shutter speeds off the starting line, when you should have been trying to freeze the wheelstand, in order to get the entire car in sharp focus.
Visceral Image wrote:
The main point is that I was able to capture a brilliant series of images with the car tracked in every frame. We're using the sequence to create a two-page spread that should look really awesome. Using 20fps now lets us choose WHICH images we want to use to show what happened. It's awesome.
you could have been using the far superior a9 electronic shutter at 20fps, back in 2017.
p.3 #18 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
OK.
osv2 wrote:
it's good that you sort of understand the manner in which your r5 is crippled, so that you don't attempt to use it where it doesn't work.
see the r5 distorted football shot below...
i agree with that, it's not much of a factor with thin-dof panning, plus you had a target-rich environment, so a couple of those pics are winners, thx for posting 'em... i've shot thousands of drag racing pics, so i appreciate seeing it.
i think that the head-on shot is a fail due to the lack of dof tho, it just doesn't work when you need to see tire wrinkle and such... you used panning shutter speeds off the starting line, when you should have been trying to freeze the wheelstand, in order to get the entire car in sharp focus.
you could have been using the far superior a9 electronic shutter at 20fps, back in 2017.
p.3 #19 · Electronic Shutter Artifacts in Sony A9 and Canon 1DX3 Images
Visceral Image wrote:
How is it a weird post? If the lights were not on, the image would be completely black...so the artificial lights were the only source of light. I provided an example where it worked just fine.
Stronger artificial lights underlying certain frequencies will inevitably cause banding in your images. There are numerous examples for all sorts of cameras, esp. with E-shutter. Only because it worked in your case doesn't mean it is perfect all the time. This neither demonstrated the others being wrong nor you being right. It is just situation dependent. If it worked in this special case, fine. It is nothing miraculous.
