Canon releases an official statement about the EOS R5 and EOS R6 heat concerns

Starting out EOS R

EOS R5 - RF24-105mm F4L, RF70-200mm f2.8L
Feb 13, 2020
295
315
tbh it looks like an emergency replacement of high speed stills shooting for very very edge cases.
Grabbed jpegs from 8K stream may and will disappoint when compared to stills in raw format.
Your probably right. I've never grabbed stills from Video so cant say what they are like and will probably do high speed stills as a primary method anyway.

I cant wait to try the R5 frame rates as although I love the R for landscape & stationary things, it drives me nuts when I try and take images of fast moving subjects even in high speed AF servo. Anything other than single shot turns out with the focus not where I want it. Maybe I'm just rubbish at panning and following a subject but the AF tracking modes on the R5 will be a game changer for me if they are as good as they look. I cant wait to get my sticky little fingers on it. :ROFLMAO:
 
  • Like
Reactions: 1 user
Upvote 0
Jan 29, 2011
10,675
6,121
You can't postprocess away the difference between sensors. If sensor A lags behind sensor B and you reduce noise on images from sensor A, you can also reduce noise on images from sensor B and it's ahead again. Postprocessing doesn't eliminate the difference and doesn't improve sensor performance.
You missed the point and you are wrong.

First off the most common examples and certainly the most consistent comparisons we see (DPReview) have been processed in an Adobe standard post process, my point was standard for two is not optimal for either. Process both optimally and the differences are not visible in normal output sizes. Putting in standard or consistent processing values for different sensors means you have to change the processing on both, not one, to get optimal results.

You are wrong because you absolutely can process away noise at the expense of detail given a higher resolution image.

Look at it like this there is a range of slider settings where the output is optimal per image and sensor, at some point during that range of different but optimal settings (visible at output sized improvement) the two sensors output overlap. You can choose either file to show more detail with less noise or the other way around, but both images look the same at output size. Now the higher density sensor image will have the noise reduction set to a higher value, but does that really mean it makes more noise, after all the lower resolution sensor will have more sharpening.

Either way it doesn't matter, the output result is visibly the same. If you don't agree show me a studio example that I can download that illustrates your point.
 
Upvote 0
You are wrong because you absolutely can process away noise at the expense of detail given a higher resolution image.

I'm right because we're dealing with the simple laws of information. Almost any processing (including noise reduction and downsampling) causes loss of information.
If sensor A and sensor B have the same resolution but different noise (A noisier than B), it's impossible to make them equal through noise reduction on A. You can de-noise A and lose information (resolution), although visually it may appear having the same noise as B. But then you can also reduce noise in B and make it even cleaner.

If A is still noisier than B but has greater resolution, you can downsample and reduce noise but again you lose information, maybe you get the same IQ as B after downsampling but it's not the original image. You can reduce both to small rectangles of 2x3 pixels (6 pixels in total) , they will look exactly the same, but what will it prove?

Look at it like this there is a range of slider settings where the output is optimal per image and sensor

I'm not sure what 'optimal' means here. I don't thing there's 'optimal' settings for any sensor. What you're saying means we're tweaking settings on A and B so that they both lose information, maybe in different ways, and eventually they look the same. Yeas maybe they may look very close but one of them will lose more information than another. So it'll never be a fair comparison.

at some point during that range of different but optimal settings (visible at output sized improvement) the two sensors output overlap. You can choose either file to show more detail with less noise or the other way around, but both images look the same at output size.

As in my previous message, I have a 45mp sensor and you have a 20mp sensor - please tune your setting up to get the same result on a chosen output image of 8192 x 5464 px.
 
Last edited:
Upvote 0
Jan 29, 2011
10,675
6,121
Now you are changing what I said to a pathetic degree. I never said you could make all sensors equal in every aspect, with a higher resolution sensor you have the option to trade noise and detail to match noise levels of lower resolution sensors, ergo there is no noise advantage to lower resolution sensors.

By optimal processing I mean processing an image file as well as you can to get the image quality you need. Simple and practical, not theoretical and irrelevant.

But show me some examples we can all download to illustrate your point that disproves mine.
 
  • Like
Reactions: 1 user
Upvote 0

dtaylor

Canon 5Ds
Jul 26, 2011
1,805
1,433
You can't postprocess away the difference between sensors.

There isn't much of a high ISO difference at the same view size.

If sensor A lags behind sensor B and you reduce noise on images from sensor A, you can also reduce noise on images from sensor B and it's ahead again.

Except sensor A is 45mp and sensor B is 20mp. Assuming a real difference in noise quantity between them, if you NR sensor B to the same extent as sensor A then sensor B is going to be quite soft and lacking in detail.

Though, again, for the same generation there isn't really a quantity difference to speak of. There is a quality difference in that high MP noise tends to come off as sharper and can therefore seem more obtrusive. But that's trivial to fix in post.
 
Last edited:
Upvote 0

dtaylor

Canon 5Ds
Jul 26, 2011
1,805
1,433
If A is still noisier than B but has greater resolution, you can downsample and reduce noise but again you lose information, maybe you get the same IQ as B after downsampling but it's not the original image.

No, it's "only" as good as B. So even if we assume greater resolution = greater noise, you're not losing anything. You've got resolution at low ISO and similar noise IQ at high ISO.

As in my previous message, I have a 45mp sensor and you have a 20mp sensor - please tune your setting up to get the same result on a chosen output image of 8192 x 5464 px.

You can't, but that's a false analogy. You're asking him to add information where non exists. Matching high ISO performance involves removing noise while retaining at least as much information as the other camera. If you have two 20mp sensors with a 1ev difference in high ISO noise it's impossible. But if you have a 45mp and a 20mp sensor with a 1ev difference in noise, it's quite doable.

But...again...the quantity differences aren't there, at least not to the extent people believe they are. There is literally not a 1ev difference between a 5Ds/sR and a 1DX mark III. Try it in DPReview's studio tool. Put it in RAW, then put the 5Ds/sR at any ISO you want and set the 1DX mark III 1 stop higher, Comp or Print viewing mode. The 1DX mark III is always worse meaning the difference between them is <1ev. Yet common wisdom says the former was horrible at high ISO and the latter is the gold standard.

And before you say 'Canon capped the 5Ds/sR at 12,800 while the 1DX mark III can go to 819,200" that is purely marketing driven, and also purely marketing drivel. 1DX mark III RAWs fall apart passed 51,200 and I would be hesitant to use any current FF body at 25,600. If you want 25,600 from a 5Ds underexpose 1ev and push in post. There are not separate amp stages for these ISOs in any shipping sensors, by these ISOs they are all pushing digitally any way.
 
Upvote 0
No, it's "only" as good as B. So even if we assume greater resolution = greater noise, you're not losing anything. You've got resolution at low ISO and similar noise IQ at high ISO.

When you downsample, you gain in noise reduction but lose reslution, overall you're still losing information.

You can't, but that's a false analogy. You're asking him to add information where non exists.

That was the point exactly. And requiring one to shrink his images is also not fair :)

Matching high ISO performance involves removing noise while retaining at least as much information as the other camera. If you have two 20mp sensors with a 1ev difference in high ISO noise it's impossible. But if you have a 45mp and a 20mp sensor with a 1ev difference in noise, it's quite doable.

That's how photonstophotos and DxO do their measurements. It's ok as long we know how it's measured and the limitations of the measurements. But I was arguing over a bit different statement on manipulating sliders for some 'optimal' results from both sensors.
 
Upvote 0
Now you are changing what I said to a pathetic degree. I never said you could make all sensors equal in every aspect, with a higher resolution sensor you have the option to trade noise and detail to match noise levels of lower resolution sensors, ergo there is no noise advantage to lower resolution sensors.

I was just illustrating that it's not that easy to 'post process away' the differences between the sensors.

By optimal processing I mean processing an image file as well as you can to get the image quality you need. Simple and practical, not theoretical and irrelevant.

But show me some examples we can all download to illustrate your point that disproves mine.

Unfortunately I only have one camera atm. We can take examples from DPR. What exactly are you suggesting to compare?
 
Upvote 0

dtaylor

Canon 5Ds
Jul 26, 2011
1,805
1,433
When you downsample, you gain in noise reduction but lose reslution, overall you're still losing information.

But not in relation to camera B.

And requiring one to shrink his images is also not fair :)

"Fair" has nothing to do with it. And we view/print images to specific sizes. If the 20mp camera has enough resolution for your view size, then so does the 45mp one scaled down to eliminate noise.

That's how photonstophotos and DxO do their measurements.

I'm looking at images, not graphs. PTP can yield valuable info, but an arbitrary threshold in a pixel S/N test is not a 2D photographic print.
 
Upvote 0

dtaylor

Canon 5Ds
Jul 26, 2011
1,805
1,433
I was just illustrating that it's not that easy to 'post process away' the differences between the sensors.

When it comes to noise you honestly don't have to post process anything. Simply viewing at the same size rather than pixel peeping will result in noise appearing very similar because it is for the same sensor size.
 
Upvote 0
Jul 19, 2011
421
283
You're not recording when you're pre-roll checks like framing, color, exposure, etc...
Therefore the camera will run all day long, or until the battery runs out. It's the active RECORDING that generates the heat. We know this factually because it's the same case with existing cameras (DSLR and MILC) when in video mode. Simply monitoring the sensor via EVF or Liveview screen takes comparably little processing power.

Magic Lantern on a 5D Mk3 is able to display the sensor temperature.
One can easily monitor how the sensor temperature increses pretty fast.
It stays at a somewhat high temperature after some time, so yes, you
can run it all day in preview mode.

But you will have an already elevated temperature on your hands when
you finally start recording. R5/R6 are no different.
 
Upvote 0
Jul 19, 2011
421
283
In fairness Canon has never positioned this camera as A-roll for a professional "video production."

Right, BUT! You could run the 5D Mk3/4 all day without ever overheating.
A few dropped frames every 12 minutes / 4GB are much more acceptable
than 20 minute waiting intervals.

This is a big step back in usability in my eyes.
 
Upvote 0
"Fair" has nothing to do with it. And we view/print images to specific sizes. If the 20mp camera has enough resolution for your view size, then so does the 45mp one scaled down to eliminate noise.

Scaling down itself is an arbitrary process as you're trying to equalise both cameras by choosing arbitrary target size. As above, at the extremities, we can have a 3x2 px rectangle which will eliminate all the differences between the sensors, and the resolution of the sensor B (higher resolution). Why are you downscaling B to A, not upscaling A to B? Maybe we need to print very large.

When it comes to noise you honestly don't have to post process anything. Simply viewing at the same size rather than pixel peeping will result in noise appearing very similar because it is for the same sensor size.

But why we choose the lower-mpix sensor as a target size?
 
Upvote 0
Grant Petty yesterday answered that en passant as well.
The new URSA 12k downsamples from a symmetrical
sensor pattern to yet unseen color excellence.

Interesting. BM sure has an interesting codec in BM BRAW. The files. though, are partly demosaiced in-camera, which probably makes these kinds of tricks possible and are a good compromise between file size and quality. I do not know the internals of the RAW format in R5, though. It seems to be Canon Cinema RAW Light, though, as the compression factor is about 1:5. (It would be cool if vendors did not call these lossy-compressed un-demosaiced or partly demosaiced formats RAW, but well, here we are and there's nothing I can do about it.)

I *guess* pulling some kind of downsampled 4k Canon Cinema RAW (light) from the 8k R5 Bayer-sensor is a bit too processor intensive for even the R5 processor. We'll see what Canon engineers come up with in the future.
 
Upvote 0

dtaylor

Canon 5Ds
Jul 26, 2011
1,805
1,433
Scaling down itself is an arbitrary process as you're trying to equalise both cameras by choosing arbitrary target size.

Within a given technological generation same format sensors are essentially equal in high ISO noise performance. They only appear unequal when you arbitrarily enlarge one more than the other, otherwise known as pixel peeping. Telling you to view at the same size and/or apply some NR before downsizing for print is not an attempt to make the sensors equal. They already are virtually equal in noise quantity. It's an attempt to get you away from pixel peeping (which is explicitly unequal), and also to explain how you can deal with the qualitative difference in noise from higher pixel density sensors (it's literally sharper).

I typically do not reference DxO because A) they are prone to error, and B) I've never cared for how they present their results (i.e. sensor scoring). But some of their tests do provide valuable insight. Do you know what their Sports test scores tell us? That high ISO differences between FF cameras are negligible. They show a 1/3rd stop difference between the 5Ds and 5D mark IV, and maybe a half stop between the 5Ds and 1DX mark III. This is in rough agreement with what I see when looking at DPReview RAW samples. And these aren't even of the same tech generation. The 5D IV and 1DX mark III have lower pixel densities and newer tech, and that's all the improvement they can deliver. You'll find the differences to be minor within both the Nikon and Sony lines as well.

Pixel density simply does not impact high ISO noise to the degree people believe. Nor is technology improving high ISO performance very quickly, if at all with some generations. The easy gains here were made in the 2000's. We're now at the point that photon shot noise dominates by a wide margin, and the electronics are about as good as they're going to get without active cooling or CFA removal. If you're expecting the R5 to be leaps and bounds better than the 5Ds/sR at high ISO, or the R6 to be better than the R5, you've set yourself up for disappointment. (Though, knowing how human beings perceive the world, if you believe R5/R6 high ISO will be dramatically better than any FF camera before, then you will likely experience that regardless of the truth.)

Why are you downscaling B to A, not upscaling A to B?

If you upscale A to B the noise quantity will be nearly identical. The noise quality will be different because everything in A will be blurrier than it appears in B.

But why we choose the lower-mpix sensor as a target size?

Because as a general rule people don't expect to make 48" prints from ISO 12,800 shots.
 
Upvote 0