The Canon EOS R3 will be 24mp, confirmed by EXIF data

HenryL

EOS R5
CR Pro
Apr 1, 2020
289
749
Nope.

I just did a quick test with R5: those resolution figures change when it's set to lower resolution.

In a full resolution image:

Focal Plane X Resolution : 5773.079634
Focal Plane Y Resolution : 5769.799366

which work out to 45 megapixels as expected.

In a M-quality jpeg:

Focal Plane X Resolution : 4093.023256
Focal Plane Y Resolution : 4088.701162

which work out to 22 megapixels.

So those particular numbers don't prove anything.

As far as I can see, the only field in the M-quality image exif that reveals the sensor is actually bigger is "Sensor Width", which is the same, 8352, in both images. I haven't looked at any of the R3 images in the web to see how it looks in them, but I'm sure somebody will. :)
Interesting, thanks. What tool are you using to view the EXIF, I don't see anything listed in the options of exifviewers.com labelled "Sensor Width".
 

Alan B

1DX Mark III
Feb 2, 2021
27
40
I still think when its announced that its going to have a 30MP or 45MP sensor!!

The one, said photographer is using at the Olympics is just a pre-production model anyway to test, so its NOT final!
 

dtaylor

Canon 5Ds
Jul 26, 2011
1,742
1,305
To answer those criticisms about the R5 not being able to operate at 20 fps f/22 and on low battery charge, I did a quick burst this morning of some dogs running around as I was out walking looking for birds with the R5+100-500mm at 500mm. Here is a gif of 53 shots in 2.71sec = 19.6 shots/sec, at f/22, 1/1000s and 26% battery charge using electronic shutter. The white dog was tracked and in sharp focus throughout.
Does the R5 keep the lens stopped down during the burst? Many lenses can't open/close the aperture fast enough to f/22 for 8-10 fps much less 20. I'm wondering if Canon has improved aperture diaphragm performance or if they just keep the lens stopped down when light is sufficient for DPAF to work.
 

dtaylor

Canon 5Ds
Jul 26, 2011
1,742
1,305
Show me what you're referring to.
You can try this comparison with any number of cameras made over the past 10 years. If both the format and the view size are the same and the tech level is comparable, then higher pixel density does not result in visibly worse noise except at the most extreme (and unusable) ISOs. This is actually what you would expect given gapless microlenses and the relative strengths of shot noise and e-noise in a modern sensor.

Screen-Shot-2021-07-28-at-12.26.05-PM.jpg
 

AlanF

Stay at home
CR Pro
Aug 16, 2012
8,691
11,706
Does the R5 keep the lens stopped down during the burst? Many lenses can't open/close the aperture fast enough to f/22 for 8-10 fps much less 20. I'm wondering if Canon has improved aperture diaphragm performance or if they just keep the lens stopped down when light is sufficient for DPAF to work.
I might be wrong, but I think in general MILCs keep their lenses stopped down for AF, unlike DSLRs, which require wide aperture for their phase detect to work. From what I recall from earlier discussions, for example, the Sony A7RIII and IV switch from initial phase detect followed by contrast detection at wide apertures to just contrast detect at narrower apertures, so they aren't opening and closing at f/8 and narrower. The R5 phase detect focusses well at narrow apertures so doesn't have to open up to focus. Somebody please correct me if I have got it wrong.
Edit - see my later post. At dtaylor’s suggestion, I checked that the initial AF is done wide open but then at 20 fps the lens remains stopped down between shots.
 
Last edited:

dtaylor

Canon 5Ds
Jul 26, 2011
1,742
1,305
The screenshot I took best illustrates this.
You're magnifying the higher resolution images more. If you normalize the size, in either direction up or down, then the higher resolution sensors are actually a bit cleaner in the comparison you chose.

You may counter that the AI processing could be working on the full image. But I highly, highly doubt that and you said the same. I won't say with certainty that they are not without a reference, but I cannot imagine that they are. The R5 overheats shooting 8k RAW and AI analysis is going to be roughly as intensive as anything involved in writing RAW video, if not more so. So how could it possibly analyze 45mp frames at a minimum of 20 Hz (would likely have to be much higher) continuously without overheating?

Everyone is probably line skipping sensor readouts at 12-bits or less for the AF processing and EVF display.

I do know that lower megapixel sensors perform better in lowlight when it comes to noise because of the pixel size.
They do not except at the very highest ISOs. Not even at 51,200 as illustrated here. You can't use different view sizes and claim that lower MP sensors perform better in low light. That's like comparing a 20x30 Velvia 50 print to a 4x6 Delta 3200 print and declaring that higher ISO films have less grain.

Screen-Shot-2021-07-28-at-12.35.12-PM.jpg
 
Aug 7, 2018
353
305
You can try this comparison with any number of cameras made over the past 10 years. If both the format and the view size are the same and the tech level is comparable, then higher pixel density does not result in visibly worse noise except at the most extreme (and unusable) ISOs. This is actually what you would expect given gapless microlenses and the relative strengths of shot noise and e-noise in a modern sensor.

View attachment 199255
If you do that comparison, you will not see a lot of difference between a new camera and a ten year old one. You will see a big difference if you download the RAWs and then overexpose them by five stops. Then the bottles look very different. Even at ISO 100 you will see a lot of noise on ten year old cameras. That will be the first test I will do once the R3 RAWs are available.
 

dtaylor

Canon 5Ds
Jul 26, 2011
1,742
1,305
I might be wrong, but I think in general MILCs keep their lenses stopped down for AF, unlike DSLRs, which require wide aperture for their phase detect to work. From what I recall from earlier discussions, for example, the Sony A7RIII and IV switch from initial phase detect followed by contrast detection at wide apertures to just contrast detect at narrower apertures, so they aren't opening and closing at f/8 and narrower. The R5 phase detect focusses well at narrow apertures so doesn't have to open up to focus. Somebody please correct me if I have got it wrong.

For some reason I was under the impression that Canon RF bodies (or at least the R) viewed at open aperture. They could be shooting continuously stopped down though. You could find out just by staring down the front of one of your lenses while playing with the AF and shutter buttons.
 

dtaylor

Canon 5Ds
Jul 26, 2011
1,742
1,305
If you do that comparison, you will not see a lot of difference between a new camera and a ten year old one. You will see a big difference if you download the RAWs and then overexpose them by five stops. Then the bottles look very different.
Agreed, but I'm talking about high ISO performance, not base ISO dynamic range. Since you bring it up though, DR is another area where people assume pixel size matters and it hasn't for quite some time. The highest scoring 35mm sensors on DR right now are also the highest density sensors.

Edit: Also, I didn't mean to imply that a current camera would show no high ISO advantage against one from 10 years ago. That's why 'comparable tech' level was one of the stated conditions. I meant to say that pixel density has not mattered much for at least the past decade. If you go further back, before gapless microlenses, then pixel density did have a significant impact on high ISO noise at the same view size. High pixel densities meant more chip space spent on circuitry and without gapless microlenses, lost photons.
 
Last edited:

AlanF

Stay at home
CR Pro
Aug 16, 2012
8,691
11,706
If you do that comparison, you will not see a lot of difference between a new camera and a ten year old one. You will see a big difference if you download the RAWs and then overexpose them by five stops. Then the bottles look very different. Even at ISO 100 you will see a lot of noise on ten year old cameras. That will be the first test I will do once the R3 RAWs are available.
That's the big difference for me. Shooting the 5DSR and even the 5DIV, I was careful to get the exposure right. My rule of thumb for BIF, for example, was to use automatic iso but overexpose by 1.7ev. However, when I got a Nikon D850 and D500 to go with the 500PF, I found with their excellent sensors I could push through few ev with just as good results without the overexposure. Now, with the R5 I have given up auto iso and use full manual and guess the exposure from the evf, underexposing not to clip highlights knowing I can push through 3 ev with no problems.
 

SteveC

R5
CR Pro
Sep 3, 2019
2,437
2,306
You're magnifying the higher resolution images more. If you normalize the size, in either direction up or down, then the higher resolution sensors are actually a bit cleaner in the comparison you chose.
And this explains your disagreement with Mike. He's looking at individual pixels, you are looking at on a per-area basis.

He's right when talking per pixel, you're right when talking about the overall picture. Thus you'll never convince him he's wrong (and vice versa), until and unless you both get on the same page.
 
  • Like
Reactions: Bahrd and pic

tron

EOS R5
CR Pro
Nov 8, 2011
4,962
1,316
Maybe Canon allowed on purpose the leak about 24Mp. In that way users will be ... psychologically prepared :LOL:

They will have cried, despaired and then more calmly will start thinking about it and liking it! Who knows? :cool:
 
  • Like
Reactions: Jack Douglas

dtaylor

Canon 5Ds
Jul 26, 2011
1,742
1,305
And this explains your disagreement with Mike. He's looking at individual pixels, you are looking at on a per-area basis.

He's right when talking per pixel, you're right when talking about the overall picture. Thus you'll never convince him he's wrong (and vice versa), until and unless you both get on the same page.
There's no reason to look at individual pixels unless the argument is that the AI AF looks at pixels in a way that makes per pixel noise matter. I don't see how/why that would be the case. Even the general assertion that image noise would present a difficulty to an AI is up for debate. I would think AI-based systems would be quite good at ignoring regular image noise so long as that was part of the initial training.
 
  • Like
Reactions: scyrene and Joules

SwissFrank

from EOS 1N to R
Dec 9, 2018
624
352
No. All that has been established is that the R3s used in the Olympics (that have pictures posted publicly) can produce 24 mp images (or images that indicate that in their exif data).

In theory, it remains possible that the sensor is bigger but the photogs with loaner R3s have agreed (as a condition to getting them, presumably) not to take or at least not publish any pics with more resolution, or that the cameras indeed have a prerelease firmware that prevents it (or fakes the exif). Not very likely, I think, but not yet proven impossible either.

I've done software for 30 years and software in Japan for over a decade. I think it's quite likely the firmware in the demos is not the release candidate, and it would have been trivial to limit the resolution had they wanted. (Note I'm not saying they did this--I don't have an opinion on actual resolution here, just on narrow question of whether pre-release bodies would nonetheless have a release-candidate firmware.)

You wouldn't want to depend on photographers with tight deadlines being sure to police every image.

If I were Canon, and Canon were releasing a 48MP camera with a 24MP temporary cripple mode, I'd let the camera take higher-res photos but have blocked the user from exporting larger files. Maybe let it export a low-res "RAW" that they could work with, or a hi-res "RAW" that Canon could unlock for them either in a service bureau or by releasing updated drivers once the resolution had been announced.
 

SteveC

R5
CR Pro
Sep 3, 2019
2,437
2,306
There's no reason to look at individual pixels unless the argument is that the AI AF looks at pixels in a way that makes per pixel noise matter. I don't see how/why that would be the case. Even the general assertion that image noise would present a difficulty to an AI is up for debate. I would think AI-based systems would be quite good at ignoring regular image noise so long as that was part of the initial training.

I tend to agree, but a lot of people don't, they pixel-peep and that colors their evaluation of everything. I get to watch this particular wrangle here again and again, each side talking past the other because they don't realize the other person is/is not pixel peeping.
 
  • Like
Reactions: scyrene

AlanF

Stay at home
CR Pro
Aug 16, 2012
8,691
11,706
For some reason I was under the impression that Canon RF bodies (or at least the R) viewed at open aperture. They could be shooting continuously stopped down though. You could find out just by staring down the front of one of your lenses while playing with the AF and shutter buttons.
I have done the experiment for you, using the EF 50mm f/1.8 STM + adapter (and learned a few things). In initlal focussing, it focusses wide open, and you can use the depth of field preview button to see the dof stopped down. And, presumably this is how it works for single shot. I then took a series of selfies at 20 fps and could see the lens remained stopped down during shooting. Interestingly, the lens stopped itself down when I was pointing it at a bright light at f/1.8.
 
  • Like
Reactions: kaihp and dtaylor