Canon R5ii vs R7ii

Alastair Norcross said:

shroob said:

Yes, that's the only advantage of the R7ii I see. However, I wouldn't want the current R7 as it's 'buggy' from what I can tell (and others in this thread have said 'a good idea executed poorly'.

Click to expand...

As others have said, this may be the opinion of some (some of whom don't actually own the camera), but it's far from the experience of many others. I think the R7 is a very well designed camera. It's easily the best crop camera Canon has ever made (and I've owned and used extensively a few, including the 7D and 7DII), and as good as the best crop cameras any other manufacturer has ever made. It was already amazing value for money at its release price ($300 less than the 7DII, released eight years earlier), and now it's an absolute steal. The AF isn't quite as good as the R6II, but it's still very impressive. People tend to remember criticisms a lot more than compliments, and the same goes for opinions about cameras. One or two persistent, and loud, critics can give the impression that a product has serious problems, when it really doesn't.

Click to expand...

Absolutely, theres always that strong influence of justifying more expensive cameras in the mix too as well as the good old crop vs fullframe battles.

I have the R5 and the R7 and generally use my R7 for smaller wildlife, unless its for low light ie nocturnal. Maybe if I had the 1.4 it would be different.

[…]

shroob said:

Yes, that's the only advantage of the R7ii I see. However, I wouldn't want the current R7 as it's 'buggy' from what I can tell (and others in this thread have said 'a good idea executed poorly'. I was hoping that the R7ii would address some of those issues e.g. more reliable autofocus, better buffer, maybe even a stacked sensor.

Click to expand...

what bugs are you seeing?

jj

Otaraka said:

Alastair Norcross said:

shroob said:

Yes, that's the only advantage of the R7ii I see. However, I wouldn't want the current R7 as it's 'buggy' from what I can tell (and others in this thread have said 'a good idea executed poorly'.

Click to expand...

As others have said, this may be the opinion of some (some of whom don't actually own the camera), but it's far from the experience of many others. I think the R7 is a very well designed camera. It's easily the best crop camera Canon has ever made (and I've owned and used extensively a few, including the 7D and 7DII), and as good as the best crop cameras any other manufacturer has ever made. It was already amazing value for money at its release price ($300 less than the 7DII, released eight years earlier), and now it's an absolute steal. The AF isn't quite as good as the R6II, but it's still very impressive. People tend to remember criticisms a lot more than compliments, and the same goes for opinions about cameras. One or two persistent, and loud, critics can give the impression that a product has serious problems, when it really doesn't.

Click to expand...

Absolutely, theres always that strong influence of justifying more expensive cameras in the mix too as well as the good old crop vs fullframe battles.

I have the R5 and the R7 and generally use my R7 for smaller wildlife, unless its for low light ie nocturnal. Maybe if I had the 1.4 it would be different.

Click to expand...

I have mentioned before that after a few years I have decided not to use the 1.4 on my R7 and 100-500. It does well and I have had good success but the keep rate goes down, especially with BIF. It tends to suffer if you crop too much. I can push cropping more without.

The 1.4 was excellent on my R5 and is also excellent on my R6II. Faster AF.

The 'bug' I referred to (maybe not the right word) is the autofocus that has dogged the camera. e.g.
I have seen friends use the R7 and they also have this problem where a few pictures will be soft as the autofocus seems to jump on and off the subject.

shroob said:

The 'bug' I referred to (maybe not the right word) is the autofocus that has dogged the camera. e.g.

I have seen friends use the R7 and they also have this problem where a few pictures will be soft as the autofocus seems to jump on and off the subject.

Click to expand...

Kinda like Zeeeeeeee explains, the R7 can be pretty darn good. It's just that the R5 IME does a bit better job (esp for BIFs, for me). Plus there's the other features that make the R5 work better for me.

I have however had some excellent success with the R7 shooting people and sports. I'll post a link to a previous thread that detailed my first big test with it, and it passed with flying colors!

https://www.dpreview.com/forums/thread/4668613

I was astounded that it only missed focus on a couple of shots out of some 1200+, in some of the most demanding conditions possible.

R2

--
Good judgment comes from experience.
Experience comes from bad judgment.
http://www.pbase.com/jekyll_and_hyde/galleries

ThrillaMozilla said:

John Sheehy said:

ThrillaMozilla said:

From your description, you need more pixels per duck. The appropriate index for that is focal length divided by pixel pitch.

Either the currently sold R7 or the R5 or R5 II with the same lens and the 1.4 extender will give approximately the same number of pixels per duck.

The aperture diameter of the lens will be the same with either alternative, so the lens will collect the same number of photons from the duck either way. Therefore, the low-light performance will be essentially the same, although different ISO settings will be used to achieve the same performance.

Click to expand...

There's a huge difference in visible read noise, though. I am not sure what drives people to want to say that we have conquered read noise and now photon noise is the only significant noise.

Click to expand...

Yeah, mea culpa for not wanting to discuss read noise, because it's a little complicated. After correcting for the different pixel pitches, assuming the same lens, the correction for pixel pitch is 0.46 log2 units, so corrected for area, the R7 curve can be adjusted upward by that amount in this comparison. But if you use the 1.4x converter with the R5 only, you can double the ISO values, i.e. shift the R5 curve to the right.

It looks like they're kind of similar overall, with the R5 a little better at low ISO and the R7 a little better at high ISO. Did I miss something here? I omitted discussion of read noise for simplicity, and because they're not vastly different.

Click to expand...

Well, yes. I don't have faith in standard deviations; they seem to be a mathematical deity to many people, but I'm just not seeing it, finding them to be fragile, and not very precise in defining the entire SNR frequency spectrum. I am fond of exposing and destroying false proxies and confounders for relevant issues, and so armed with the knowledge of what happens with mosaiced data and some of the common sensor readout artifacts, I once created a totally random-Gaussian synthetic black frame and another with the perfect storm of spatially-correlated events seen in real sensors, and adjusted the intensity so that they had the same standard deviation, and the difference was tremendous. I could downsample the random Gaussian enough so that it first appeared completely black, and downsampling the perfect storm version to the same size was still a rainbow of bold colored noise, with both going through exactly the same processing. This was the biggest "Black Swan" event that made me aware that single-value standard deviations of sensor noise had no direct correlation to to the real issue with noise; its visibility.

Recently I have been binning black frames from a few different cameras, using a random Gaussian reference, too, and all the sensors have more noise than they should at lower (more visible) frequencies, but they vary quite a bit. For example, The R5 has slightly more DR at ISO 100 than the R5-II, but if you bin black frames from both, as you get to 4x4 or so, and even larger bins, the R5-II has the greater DR! Similar spectral differences happen when I do it with the R7 vs the R5; at the pixel level, the R7 has almost a stop less DR than the R5, but as the bins get larger, they pull closer, to the point that the R7 actually has just as much DR per unit of sensor area. When I bin in only one dimension, banding noise is also independent of original pixel standard deviations.

So, IMO, taking pixel level standard deviations and then extrapolating them to visible noise is wrong; you should actually bin the pixels, rather than assume what they should bin to, based on truly uncorrelated noise, which DOES NOT EXIST in real sensor read noise.

BTW, P2P's "input-referred" is input to the photosites that actually result in electron charge; not photons coming from the lens, so exposure is separated from electron charges by the quantum efficiency.

ThrillaMozilla said:

ThrillaMozilla said:

Here's how most people seem to compare noise (same ISO and exposure, at 1:1 pixel view),

Click to expand...

I agree, that's the wrong way to compare. People fool themselves without realizing it.

Click to expand...

I have often marveled over the likelihood of someone who "trusts their eyes" receiving a prototype 2 gigabyte camera with 100% true QE and no read noise, looking at 100% pixel views on a 100 PPI monitor, and saying, "yuck; this sensor ruins everything with ridiculous softness and noise".

ThrillaMozilla said:

ThrillaMozilla said:

and even here, the R7 holds its own with read noise considering the range of pixel sizes with 3.46x the area with the A9-III, and 1.84x with the R5 bodies:

100%

The Comparison tool doesn't make it possible to compare with the same total light, but we get as close as we can, leaving the R7 with 1/3 stop less total light (exposure times area):

Almost the same total light

So, the R7 is better, even with 1/3 stop less total light.

Click to expand...

The visual between R5 and R7 sort of agrees with my calculation, although I don't see much difference between the R5 and R7 in your second, more realistic example. If R7 is better in that demo -- and I'm not sure it is -- I don't see it as much better.

Click to expand...

Maybe you're thinking in a way where all absolutely-bad-looking things are basically the same and not worth ranking, but the portrait's coarser features punch through the noise best with the R7, and that is with the R7 receiving 1/3 stop less total light. If it received the same light as the FF sensors, it would punch through even better.

The R7's sensor and its implementation is clearly the best, per unit of sensor area, in the Canon sensor stable. The R6 has similar-looking noise, but has less resolution, even if you use the entire frame.

John Sheehy said:

ThrillaMozilla said:

John Sheehy said:

ThrillaMozilla said:

From your description, you need more pixels per duck. The appropriate index for that is focal length divided by pixel pitch.

Either the currently sold R7 or the R5 or R5 II with the same lens and the 1.4 extender will give approximately the same number of pixels per duck.

The aperture diameter of the lens will be the same with either alternative, so the lens will collect the same number of photons from the duck either way. Therefore, the low-light performance will be essentially the same, although different ISO settings will be used to achieve the same performance.

Click to expand...

There's a huge difference in visible read noise, though. I am not sure what drives people to want to say that we have conquered read noise and now photon noise is the only significant noise.

Click to expand...

Yeah, mea culpa for not wanting to discuss read noise, because it's a little complicated. After correcting for the different pixel pitches, assuming the same lens, the correction for pixel pitch is 0.46 log2 units, so corrected for area, the R7 curve can be adjusted upward by that amount in this comparison. But if you use the 1.4x converter with the R5 only, you can double the ISO values, i.e. shift the R5 curve to the right.

It looks like they're kind of similar overall, with the R5 a little better at low ISO and the R7 a little better at high ISO. Did I miss something here? I omitted discussion of read noise for simplicity, and because they're not vastly different.

Click to expand...

Well, yes. I don't have faith in standard deviations; they seem to be a mathematical deity to many people, but I'm just not seeing it, finding them to be fragile, and not very precise in defining the entire SNR frequency spectrum. I am fond of exposing and destroying false proxies and confounders for relevant issues, and so armed with the knowledge of what happens with mosaiced data and some of the common sensor readout artifacts, I once created a totally random-Gaussian synthetic black frame and another with the perfect storm of spatially-correlated events seen in real sensors, and adjusted the intensity so that they had the same standard deviation, and the difference was tremendous. I could downsample the random Gaussian enough so that it first appeared completely black, and downsampling the perfect storm version to the same size was still a rainbow of bold colored noise, with both going through exactly the same processing. This was the biggest "Black Swan" event that made me aware that single-value standard deviations of sensor noise had no direct correlation to to the real issue with noise; its visibility.

Recently I have been binning black frames from a few different cameras, using a random Gaussian reference, too, and all the sensors have more noise than they should at lower (more visible) frequencies, but they vary quite a bit. For example, The R5 has slightly more DR at ISO 100 than the R5-II, but if you bin black frames from both, as you get to 4x4 or so, and even larger bins, the R5-II has the greater DR! Similar spectral differences happen when I do it with the R7 vs the R5; at the pixel level, the R7 has almost a stop less DR than the R5, but as the bins get larger, they pull closer, to the point that the R7 actually has just as much DR per unit of sensor area. When I bin in only one dimension, banding noise is also independent of original pixel standard deviations.

So, IMO, taking pixel level standard deviations and then extrapolating them to visible noise is wrong; you should actually bin the pixels, rather than assume what they should bin to, based on truly uncorrelated noise, which DOES NOT EXIST in real sensor read noise.

BTW, P2P's "input-referred" is input to the photosites that actually result in electron charge; not photons coming from the lens, so exposure is separated from electron charges by the quantum efficiency.

Click to expand...

Well, yes, it's all a bit imperfect, isn't it? And we have to assume that the tests are done right, even though some of them are not. Nevertheless, the standard deviations plus visuals plus limited noise spectra are all most of us have, and contrary to your experience, I usually observe reasonable correspondence between standard deviations and the DPR visual tests -- although some sensors do show visually prominent autocorrelated noise (fixed pattern, smoothing, banding, etc.).

Having said all that, the R7 sensor looks just fine to me, based on all the tests I've seen.

By the way, I don't trust the tungsten-illuminated tests in your examples, because there's a strong illumination gradient. It looks like they illuminated the target from only one side, at least for one of those cameras. For daylight illumination exposure times and illumination have been adjusted differently for the various cameras, which also makes comparison more difficult.

John Sheehy said:

John Sheehy said:

John Sheehy said:

Here's how most people seem to compare noise (same ISO and exposure, at 1:1 pixel view),

Click to expand...

I agree, that's the wrong way to compare. People fool themselves without realizing it.

Click to expand...

I have often marveled over the likelihood of someone who "trusts their eyes" receiving a prototype 2 gigabyte camera with 100% true QE and no read noise, looking at 100% pixel views on a 100 PPI monitor, and saying, "yuck; this sensor ruins everything with ridiculous softness and noise".

Click to expand...

Now there's a serious idea for a reference image. A perfect image--yuck.

John Sheehy said:

John Sheehy said:

John Sheehy said:

and even here, the R7 holds its own with read noise considering the range of pixel sizes with 3.46x the area with the A9-III, and 1.84x with the R5 bodies:

100%

The Comparison tool doesn't make it possible to compare with the same total light, but we get as close as we can, leaving the R7 with 1/3 stop less total light (exposure times area):

Almost the same total light

So, the R7 is better, even with 1/3 stop less total light.

Click to expand...

The visual between R5 and R7 sort of agrees with my calculation, although I don't see much difference between the R5 and R7 in your second, more realistic example. If R7 is better in that demo -- and I'm not sure it is -- I don't see it as much better.

Click to expand...

Maybe you're thinking in a way where all absolutely-bad-looking things are basically the same and not worth ranking, but the portrait's coarser features punch through the noise best with the R7, and that is with the R7 receiving 1/3 stop less total light. If it received the same light as the FF sensors, it would punch through even better.

The R7's sensor and its implementation is clearly the best, per unit of sensor area, in the Canon sensor stable. The R6 has similar-looking noise, but has less resolution, even if you use the entire frame.

Click to expand...

I'd have to see more carefully done tests (or do them myself) to be certain that the R7 is the best, but based on what I have seen, it looks about as good or maybe better than the other Canons. I don't have the FF others to compare with.

Meanwhile, for bird and wildlife photography, focusing speed and accuracy are the new unknowns for cameras.

John Sheehy said:

There's a huge difference in visible read noise, though. I am not sure what drives people to want to say that we have conquered read noise and now photon noise is the only significant noise.

Click to expand...

We haven't conquered read noise. It is just the shot noise already spoils images at exposures where the read noise is not an issue. No need to zoom into shadows at ISO 51200 to demonstrate the presence of the read noise, when mid-tones already look rubbish at ISO 12800 because of the shot noise.

John Sheehy said:

Here's how most people seem to compare noise (same ISO and exposure, at 1:1 pixel view), and even here, the R7 holds its own with read noise considering the range of pixel sizes with 3.46x the area with the A9-III, and 1.84x with the R5 bodies:

100%

The Comparison tool doesn't make it possible to compare with the same total light, but we get as close as we can, leaving the R7 with 1/3 stop less total light (exposure times area):

Almost the same total light

So, the R7 is better, even with 1/3 stop less total lilight.

Click to expand...

Sure, the read noise is clearly different. Neither shot is a keeper though.

Is there a hypothetical scenario where having zero read noise would have made the image usable but having 1e input referred read noise rendered the image unusable?

Dem Bell said:

John Sheehy said:

There's a huge difference in visible read noise, though. I am not sure what drives people to want to say that we have conquered read noise and now photon noise is the only significant noise.

Click to expand...

We haven't conquered read noise. It is just the shot noise already spoils images at exposures where the read noise is not an issue. No need to zoom into shadows at ISO 51200 to demonstrate the presence of the read noise, when mid-tones already look rubbish at ISO 12800 because of the shot noise.

Click to expand...

So what you say is that while we haven't technically conquered read noise photographically it never will become an issue and thus we have photographically conquered read noise.

charlyw64 said:

Dem Bell said:

John Sheehy said:

There's a huge difference in visible read noise, though. I am not sure what drives people to want to say that we have conquered read noise and now photon noise is the only significant noise.

Click to expand...

We haven't conquered read noise. It is just the shot noise already spoils images at exposures where the read noise is not an issue. No need to zoom into shadows at ISO 51200 to demonstrate the presence of the read noise, when mid-tones already look rubbish at ISO 12800 because of the shot noise.

Click to expand...

So what you say is that while we haven't technically conquered read noise photographically it never will become an issue and thus we have photographically conquered read noise.

Click to expand...

I would not say "never". I am saying that the shot noise has not been an issue for me in last 5 years or so and I'm struggling to see when it can "ruin" the image that is not already ruined by the shot noise. I hope someone will disagree.

There might be applications like astrophotography that might benefit greatly from further reduction of shot noise. I don't know.

Dem Bell said:

John Sheehy said:

There's a huge difference in visible read noise, though. I am not sure what drives people to want to say that we have conquered read noise and now photon noise is the only significant noise.

Click to expand...

We haven't conquered read noise. It is just the shot noise already spoils images at exposures where the read noise is not an issue.

Click to expand...

At high display magnification, yes. Not all photography needs high display magnification, though.

Dem Bell said:

No need to zoom into shadows at ISO 51200 to demonstrate the presence of the read noise, when mid-tones already look rubbish at ISO 12800 because of the shot noise.

Click to expand...

Again, a display magnification thing.

Dem Bell said:

Dem Bell said:

Here's how most people seem to compare noise (same ISO and exposure, at 1:1 pixel view), and even here, the R7 holds its own with read noise considering the range of pixel sizes with 3.46x the area with the A9-III, and 1.84x with the R5 bodies:

100%

The Comparison tool doesn't make it possible to compare with the same total light, but we get as close as we can, leaving the R7 with 1/3 stop less total light (exposure times area):

Almost the same total light

So, the R7 is better, even with 1/3 stop less total lilight.

Click to expand...

Sure, the read noise is clearly different. Neither shot is a keeper though.

Click to expand...

Of what value is your absolute aesthetic assessment to someone who has to photograph with low light and high shutter speeds, and given that QE varies over a fairly narrow range, the differences all lie in the visible read noise differences between sensors?

Dem Bell said:

Is there a hypothetical scenario where having zero read noise would have made the image usable but having 1e input referred read noise rendered the image unusable?

Click to expand...

Again, back to display magnification. If you take a photon-noise-only image, you can always reduce it in displayed size, such that noise is not visible. All read noise has spatial correlation, and the visibility of the noise is fractal-like, and as you reduce display size and exposure at the same time, noise remains visible. There is much more noise energy that is spatially structured in read noise, at very low image frequencies.

Thanks John.

I agree that read noise can make or break the image if the required output is heavily downsampled (say 1 Mpx for web sharing) AND if exposure was extremely limited (say ISO 51200 or above was required to get the tones right).

It is subjective where people draw the line for what is "good enough". I think I drew mine above the point where read noise affects image quality:

• at least 6 Mpx output, at least ISO 25600 (ff equivalent)
• for the 12 Mpx output, at least ISO 12800 (ff equivalent)
• for the 24 Mpx output, at least ISO 6400 (ff equivalent)
• for the 48 Mpx output, at least ISO 3200 (ff equivalent)

I'm using the ISO setting as the indicator of exposure assuming no adjustment to SOOC JPEG image lightness would be required in post.

BobKnDP said:

SvO2 said:

If you're cropping anyway, any noise/dynamic range advantage the FF sensor had is negated. And even a teleconverter causes you take a hit to the f value, so you'd have to increase the ISO (and therefore noise) to compensate regardless.

(snip)

Click to expand...

Could you explain the relationship between cropping and SNR/dynamic range?

Seems like it ought to be the same for 0.39 (1/1.6^2) of a frame as for the whole thing, assuming that no bizarre definition is used.

I have no idea whether there is some change when using crop mode with an R5.

Click to expand...

No difference at the pixel level has been noted, that I am aware of, when the R5 is in crop mode vs FF mode; the same bit depth and digitization rate is used in both cases. Crop mode simply never shows you most of the sensor pixels. Pixel-level DR is what you seem to be conceptualizing when "DR" is mentioned, but the bottom of that DR is not a solid floor below which nothing is recorded. IOW, it's not like raw highlight clipping happens at 1400DN, and black would be 0, but anything under 20DN is recorded at 0DN or 20DN. If that was how things worked, then cropping would have no effect on even final image-level DR. The bottom of DR, in the way we rate raw file DR, is the fuzzy, abstract presence of noise, and therefore, even though we enumerate dynamic ranges, the apparent "bottom" changes with magnification changes, the same way that when you view an image on-scene it gets noisier and noisier as you view at 25% to 50% to 100% to 200% to 300%, etcetera, which simulates cropping harder, but still using the same display area (the screen or window).

I own R6ii and R5. Of them two I prefer the R5 with my 100-500mm, for the extra pixels. R6ii seems the better when focusing in lower light and recovers details better but R5 is not far behind. I've set the back button to switch between full and crop mode. With small birds, I'm in crop mode most of the time. With bigger birds I'm starting in crop mode and as I'm getting closer I'm switching to full mode.

I've found R7+100-500mm limited to good light conditions while R5/R6ii a bit more tolerant to cloudy weather or the forest darkness.

I would wait for the R5ii.