Canon R6 Mark III High ISO and Dynamic Range – Good, but not Class Leading

[AlanF]

i looked but i am disappointed with the test procedure. is this really seeing a RF50 1.2L vs RF28-70 2.8L @ F4+ comparison? if so i am impressed honestly that the RF28-70 competes very favorably. to bad they didnt show the RF50 1.2L at 2.8 also. is the idea really that noise reduction in the raw files significantly impact camera resolution? if so i would advise them to find a high resolution lens they can adapt to any major system and manually focus it. as it is i see a fun curve peaking at F/4 and tailing off.

The relevant comparisons for measuring the resolution of the sensors are in the sections "Matrix Resolution". Their description of the procedure shows how carefully they made the measurements and why they didn't show f/2.8. They make it clear they measure at apertures where the lenses are limited by diffraction only and not by chromatic aberration. That is why the RF28-70 competes so well with the RF 50/1.2 - the results are independent of lens and depend only on f-number. So don't be disappointed with the test procedure, it is excellent.

Here is the section that describes this.
"We determine sensor resolution based on the MTF50 function, and measurements are typically taken on unsharpened RAW files, which we previously convert to TIFF format using dcraw. To avoid optical aberrations, we measure MTF50 values only for the f/4.0–f/16 aperture range, where diffraction is the main limiting factor. It's also worth remembering that we take between a dozen and several dozen shots at each aperture (with both autofocus and manual focus), then select the best ones. In this part of the test, in addition to the Canon RF 24-70mm f/2.8L IS USM lens, we also used the RF 50mm f/1.2L USM lens. The highest results are presented in the chart below."

 

[GMCPhotographics]

i looked but i am disappointed with the test procedure. is this really seeing a RF50 1.2L vs RF28-70 2.8L @ F4+ comparison? if so i am impressed honestly that the RF28-70 competes very favorably. to bad they didnt show the RF50 1.2L at 2.8 also. is the idea really that noise reduction in the raw files significantly impact camera resolution? if so i would advise them to find a high resolution lens they can adapt to any major system and manually focus it. as it is i see a fun curve peaking at F/4 and tailing off.

I think you will find that those lenses when stopped down far out resolve any current sensor being tested. MFT charts generally are for wide open results, most lenses resolve significanly higher when stopped down by 2 stops, usually resolving at their optical peak performance. These lenses choices allow a comparison between sensor brands using inter-brand adapters. Having 2 consistnet lenes allow us to dail out the lens performance as it's a common standard with these tests.
What is interesting, the Sony A7 IV with it's 33mp sensor scores a chart value of (as seen here: R6ii test chart) 72/73. The R5ii is scraping a low 76. The R6ii comes in at a credible 64. The original R5, scores 81/82. Which is in line with other 45mp sensors. I supsect that the R6iii when it's tested will resolve similar resolution / detail levels of the current R5ii. This makes the R6iii the detail resolution sleeper of the range, where the R5ii's main advantage is the Electronic shutter / Stacked sensor / 14bit processing. But in most other areas, the R6iii is similar or equal.
I'm not dissing the R5ii, it's an amazing camera, but these are the tested facts...there's something funky going on with the R5ii's resolved detail compared to the original R5. it's almost Like Canon have put a stronger AA filter over the sensor.

 

[ashmadux]

What is interesting, the Sony A7 IV with it's 33mp sensor scores a chart value of (as seen here: R6ii test chart) 72/73. The R5ii is scraping a low 76. The R6ii comes in at a credible 64. The original R5, scores 81/82. Which is in line with other 45mp sensors. I supsect that the R6iii when it's tested will resolve similar resolution / detail levels of the current R5ii. This makes the R6iii the detail resolution sleeper of the range, where the R5ii's main advantage is the Electronic shutter / Stacked sensor / 14bit processing. But in most other areas, the R6iii is similar or equal.
I'm not dissing the R5ii, it's an amazing camera, but these are the tested facts...there's something funky going on with the R5ii's resolved detail compared to the original R5. it's almost Like Canon have put a stronger AA filter over the sensor.

I personally have been complaining and somewhat raging about this for a while now. I was very disappointed with the r62 in terms of the image performance and the body styling and feel. That said, there's something really really weird about the resolving of the r62 sensor, and I retired the camera after a month.

One of the most basic or toughest tests that I put it through was simply taking a picture of a lamppost in the park, with ragged trees/leaves around; so there's random detail and color everywhere. But I just couldn't get sharp images out of this basic test. When I later got my R5 it was the first thing I tested and it nailed it.

I'm a certified pixel peeper that has been tracking my sensor performance since my very first camera. It's just the nature of how I work very close. I use my photos in a ton of design layouts and so I often have to use them at 100% or close.

Anyways, I noticed how r62 files, especially in the daytime, would 'look' sharp but still not sharp. It honestly drove me crazy. It did take me awhile to get used to the autofocus but then even that was wigging out. But regardless I still have a month's worth of pictures in Japan to pixel peep and scratch my head. The smoothness is just looks weird when you know something is supposed to be much sharper. And I can directly compare it to the 5d3 files which do not exhibit this behavior and can produce some razor sharp portrait images regardless of eyes or skin etc.

And on the other side of this coin, because of the nature of that smoothing or noise reduction or less resolution combination also with the AA filter, r62 night photos of especially areas of Tokyo can look absolutely amazing. It just smoothes out in all of the right ways while keeping sharp edges. I have images of the previous Gundam at night and center Tokyo and some of it just looks absolutely fantastic... Almost miraculously clean. But man the daytime photos just really grinded my gears. My portraits as well, zooming into the eyes/eyelashes have this kind of soft sharpness effect, and I hated it.

The last thing I'll say is that from the reviews of the r52, especially from the raw files samples, I could spot the noise immediately. Whether that turns into a debate for a lot of people who just didn't care or yada yada, I noticed back then that the image quality took a hit. But all it did was make me feel great about my R5 purchase.

Sometimes it's good to feel good that i didn't need to upgrade. Because there's always caveats, big or small. At least with Canon. But more noise? Absolutely not, that's a automatic no purchase for me.

Cheers all.

 

[erader]

Thanks! Very interesting to see this. Overall quite positive and not a crazy sacrifice for the pixels gained. Super happy to see this, and maybe with the R6 iv they'll keep the pixels but bump the ISO back up -- which would be a killer combo for this tier.

That stated, I think that the first comparison must always be pixel to pixel -- not downsized. If I buy a matrix of pixels then I want to use all of those pixels -- downsizing is a silly comparison. If I downsize of my R6 20mp images to 10mp they're amazing even at 52k ISO -- but why would I do that? I wouldn't. Crop, yes, but downsize for quality? Nope. I use all of those pixels to the best that I can and each pixel must stand on its own.

yeah when you hear that downsize argument you know it's noisier

 

[GMCPhotographics]

I personally have been complaining and somewhat raging about this for a while now. I was very disappointed with the r62 in terms of the image performance and the body styling and feel. That said, there's something really really weird about the resolving of the r62 sensor, and I retired the camera after a month.

One of the most basic or toughest tests that I put it through was simply taking a picture of a lamppost in the park, with ragged trees/leaves around; so there's random detail and color everywhere. But I just couldn't get sharp images out of this basic test. When I later got my R5 it was the first thing I tested and it nailed it.

I'm a certified pixel peeper that has been tracking my sensor performance since my very first camera. It's just the nature of how I work very close. I use my photos in a ton of design layouts and so I often have to use them at 100% or close.

Anyways, I noticed how r62 files, especially in the daytime, would 'look' sharp but still not sharp. It honestly drove me crazy. It did take me awhile to get used to the autofocus but then even that was wigging out. But regardless I still have a month's worth of pictures in Japan to pixel peep and scratch my head. The smoothness is just looks weird when you know something is supposed to be much sharper. And I can directly compare it to the 5d3 files which do not exhibit this behavior and can produce some razor sharp portrait images regardless of eyes or skin etc.

And on the other side of this coin, because of the nature of that smoothing or noise reduction or less resolution combination also with the AA filter, r62 night photos of especially areas of Tokyo can look absolutely amazing. It just smoothes out in all of the right ways while keeping sharp edges. I have images of the previous Gundam at night and center Tokyo and some of it just looks absolutely fantastic... Almost miraculously clean. But man the daytime photos just really grinded my gears. My portraits as well, zooming into the eyes/eyelashes have this kind of soft sharpness effect, and I hated it.

The last thing I'll say is that from the reviews of the r52, especially from the raw files samples, I could spot the noise immediately. Whether that turns into a debate for a lot of people who just didn't care or yada yada, I noticed back then that the image quality took a hit. But all it did was make me feel great about my R5 purchase.

Sometimes it's good to feel good that i didn't need to upgrade. Because there's always caveats, big or small. At least with Canon. But more noise? Absolutely not, that's a automatic no purchase for me.

Cheers all.

What are you comparing your R6II with? I’ve been really pleased with the sharpness and detail I’m seeing with my copy and my lenses. It’s far superior to my previous cameras.

 

[DocInfoSci]

You aren't paying x dollars per megapixel, you're paying for the camera on the whole.

True that. I'm looking at this as an existing R6 owner who bought the camera for the pixel quality as-is, so for the most part the interesting part is the sensor itself -- the rest is mildly interesting. I'm not "upgrading" from an R8, or a DSLR, etc. Nor would I be "downgrading" from an R1 or R5. Body robustness, vendor SLA, customizability, etc. all have their part in total value. I think the R6 III is overall a great total value, based on what I'm reading.

In this case, objectively, just downsizing is the worst-case scenario, as I mentioned in the article.

Yes, you did mention that — I saw your later remark about downsizing. It was very interesting to note.

Honestly, I was partially grumbling out loud in the background of my day job when I saw the downsizing remark — and many other times in the forums people have been like, "well I just use my ($2K more expensive camera) to down size and then I'm good to go if I need that quality." And, hey — that's not wrong. But I think that attitude mostly misses the point about whether the sensor for the R6 line moved forward, held steady, or lost capability in one or more ways; let alone the fact that many people making the move will not be gaining more in terms of body (not much changed from the R6 -> R6 III in the body department) but rather will be purchasing for more megapixels — so downsizing an image just to reclaim their prior capability in noise (etc.) seems to miss the point of the purchase for that crowd. I mean, the R6 -> R6 II was effectively the same quality in all regards for more pixels and without the need to downsize. Win all around.

I'm also an engineer and scientist responsible for the day to day operation of an analytics company focused on advanced medical research at the cellular level. How a comparison happens, and what that comparison informs, really matters to my teams. For most people it probably is a "close enough" situation.

You did a nice job with the article, and it was the first such piece that answered many of the questions that I had about the new sensor. In the context of what I mentioned above, the fact that downsizing is the worst case scenario was very interesting to me: contrary to the oft-heard chant of "just downsize to get the same quality" that does not seem to apply here -- downsizing doesn't get a person back to square one in all quality factors if the "one" was the prior R6 or R6 II. And Canon did leave the R6 II as an option. People get a real vote here with that choice of R6 II vs R6 III when both are left on the market. Aside from a desire to clear old stock, I wonder if Canon is watching for the hint of any other purchasing preference / attitude while both remain.

Thank you and the team for all of the interesting reads of late on histories, capabilities, etc.

 

[David_B]

Canon EOS R# FF DR... not the R6M3... there's not much between them, a bee's dick if you would... except for the r1 at ISO 400-1600, what happened there?

 

[David_B]

And read noise...

 

[neuroanatomist]

except for the r1 at ISO 400-1600, what happened there?

Bill Claff relies on user-submitted RAW files to generate his data. I strongly suspect that whoever sent him the R1 files used electronic shutter instead of manual shutter for the captures. Notice how with the other recent(ish) R bodies (R3, R5, R5II, R6, R6II) there are plots for both mechanical (unlabeled) and electronic (ES) shutter, but for the R1 there is only one entry and the PDR is lower than expected. But it's pretty much spot-on with the R5II in ES.



When his R1 data came out lacking the mechanical vs ES comparison, I offered to send him file sets but didn't hear back. I'll contact him again, (by email though we are practically neighbors...he lives in the next town over from me). Or maybe I'll just shoot the files per his specs and send them.

 

[David_B]

Reflecting on the above, if the R6III fits on those graphs where expected, a conclusion one might arrive at is that Canon's sensor design has evolved to deliver roughly the same IQ regardless of pixel size, meaning they could potentially go to 60 or 80 or 100MP with no loss in IQ.

 

[ddixon]

FYI - for those that use DXO PhotoLab, they have now posted an update that supports the R6mIII.

 

[jamtra]

FYI - for those that use DXO PhotoLab, they have now posted an update that supports the R6mIII.

Finally!!