Canon to release a 100mp EOS R system camera next year [CR2]

[Sporgon]

Western Digital 14TB external drives are now under $300. While there's still a good argument for the performance issues dealing with larger files especially on older computers or with older software, the cost of storage really isn't a big deal these days.

In my case I certainly value my time more than the cost of the storage.

I presume I now have shares in Western Digital

 

[justaCanonuser]

Cue "no-one needs 100megapixels, because I don't need it and my computer is too slow" comments in 3...2...1....

Nah, small pixel = low DR (limited photon collection capacity). Sharpness on pixel level only with a lot of light and very fast shutter speeds. Quick kick-in of diffraction blur at smaller apertures. All those limitations are ruled by physics. But don't be worried, its images will be not worse than a those of a 20 MP camera in many standard settings (based on much bigger files, of course).

 

[Michael Clark]

It doesn't change anything. The "advanced options" only allow you to ask will this be acceptable at X arbitrary print size? The relationship between the two (1.6x more DoF at 1.6x the distance) remains the same. Note: their calculator has a rounding error somewhere. You can get it to report 2x the DoF or 1x with extreme print sizes. You can also create situations where the reported Total Depth of Field value does not match the difference in the reported near/far values.

Anyone who thinks the sum of two rounded numbers is more correct than the sum of the actual extended numbers that is then rounded after adding them together doesn't understand what a rounding error is.

Is one-third plus two-thirds more accurately expressed as:

0.33 + 0.66 = 0.99

or

0.3333333333333333 + 0.6666666666666666 = 0.9999999999999999

which is then rounded to 1.00?

 

[Michael Clark]

Yes well there's some truth in that.

However it raises an interesting question. The 5 series was always the GP camera model that was good for most applications, and an awful lot are used to photograph weddings, the FF giving such an advantage in dimly lit churches. If the R5 really is the 5DIV's replacement then it's jumped to 45mp, and the difference between 45 and 50mp is basically nothing in both terms of resolution, output size and processing speed ! So 5 series users, often shooting hundreds of images at a time, are now 'lumbered' with 45mp data to deal with. Maybe CRAW is coming to the rescue here ? The problem with the previous MRAW & SRAW was that only DPP seemed to be able to convert them properly and even then there are issues so it wasn't really a viable option for many, myself included.

So it looks to me as if 5 series photographers are going to be forced into becoming data masochists in the future !

Or they can use the R6 instead... at least for many parts of the event.

 

[Michael Clark]

Western Digital 14TB external drives are now under $300. While there's still a good argument for the performance issues dealing with larger files especially on older computers or with older software, the cost of storage really isn't a big deal these days.

I just picked up a 12TB WD Elements external HD for 180+tax at New Egg. That's only $15/TB. It was a one day sale. That still doesn't mean I like needing an entire closet just to store all of my archival drives. (And another at an offsite location if I'm being as careful as I should.)

 

[jolyonralph]

People almost always look at these things from their own perspectives and don't fully embrace other peoples perspectives.

The cost of storage for many is comparatively modest, but for high volume shooters it can very quickly get out of hand. I know of a guy who shoots with 1 series cameras and sells big prints amongst a lot of regular print and digital output. His lowest shutter count body, of three, is over 300,000 in three years, his highest was well over 400,000. So he is at 1,000,000 RAW files in three years just off his main bodies, he often uses remotes and second and third shooters. I know of one event where he had 15 cameras shooting over three days.

I worked for an event shooting company a couple of times and they used ten to fifteen photographers in different sets and expected groups to be staged and shot within five minutes. They would be processing at least 3,000 shots an evening, over a three day event they might be processing 20,000 images.

But those are a couple of pro situations, now the R5 can shoot 45mp at 20fps even amateur wildlife and sports shooters can rack up thousands of images in a day.

I expect HD storage costs are pretty low on the list of his expenses though even for that amount of photography.

A million photos is around about 25TB, or at most $1200 a year in HD storage costs assuming you mirror everything onto two drives for safety. A bit more if you have a RAID system, but that cost would be spread over multiple years.

The cost of HD storage is falling faster than the rise in raw file sizes

 

[jolyonralph]

Nah, small pixel = low DR (limited photon collection capacity). Sharpness on pixel level only with a lot of light and very fast shutter speeds. Quick kick-in of diffraction blur at smaller apertures. All those limitations are ruled by physics. But don't be worried, its images will be not worse than a those of a 20 MP camera in many standard settings (based on much bigger files, of course).

Then explain the DR performance of the A7RIV please

 

[dtaylor]

Anyone who thinks the sum of two rounded numbers is more correct than the sum of the actual extended numbers that is then rounded after adding them together doesn't understand what a rounding error is.

Did you even read what I wrote? Because your snarky commentary has nothing to do with what I described or witnessed.

 

[dtaylor]

From further into the same Bob Atkins article you cited:

"Again, this simple analysis only applies at "intermediate" distances, but we have to have that limitation if we want a "simple" formula. It only really breaks down when the lens is focused further than about halfway to the hyperfocal distance or when we get to magnifications near 1:1"

Photograph a focus test chart with your 5D4 and 7D2, same lens and aperture, back up to preserve framing. See what you get.

 

[dtaylor]

Nah, small pixel = low DR (limited photon collection capacity).

The highest DR 35mm sensors at DxO and PtP are high pixel density sensors. At DxO the top 9 spots are held by high density sensors.

Sharpness on pixel level only with a lot of light and very fast shutter speeds.

This has not been my experience at 50mp, and I don't hear 90D or M6 mark II owners complaining at even higher pixel density.

Quick kick-in of diffraction blur at smaller apertures.

And for a few stops beyond that you'll still see IQ gains.

 

[Michael Clark]

Photograph a focus test chart with your 5D4 and 7D2, same lens and aperture, back up to preserve framing. See what you get.

you do it at macro (1:1) distances and see what you get.

 

[Michael Clark]

I expect HD storage costs are pretty low on the list of his expenses though even for that amount of photography.

A million photos is around about 25TB, or at most $1200 a year in HD storage costs assuming you mirror everything onto two drives for safety. A bit more if you have a RAID system, but that cost would be spread over multiple years.

The cost of HD storage is falling faster than the rise in raw file sizes

Cost of storage media isn't everything. The time to properly manage all of that data is the real clincher. You can't just copy files to an archive drive and bury it. You must periodically check the drive, and the files on it, to insure they are still viable. Otherwise there's no point in considering them a backup.

 

[Michael Clark]

I expect HD storage costs are pretty low on the list of his expenses though even for that amount of photography.

A lot of folks expect they understand all facets of issues that they don't deal with themselves. They think they know all of the answers when they don't even have a clue what half of the questions are.

How many customers do you have who expect that as part of what they paid you for photographic services, you will maintain copies of the images you produced for them?

How many customers have a contract with you that specifically states that you will maintain such copies and provide them if needed for a certain period of time? One year? Three years? Five years? Ten years?

 

[Michael Clark]

Did you even read what I wrote? Because your snarky commentary has nothing to do with what I described or witnessed.

That's exactly what you wrote.

You claimed Cambridge's DoF calculator is inaccurate because a number rounded from the end result of an equation that used numbers extended to many more significant digits during all of the calculations does not agree with the sum of two numbers that were rounded before being added together. Anytime one uses a rounded number at a step in a calculation, the result becomes less accurate. When one adds two rounded numbers together, the sum of those two numbers has the potential to compound the variation by up to 2X as much as the variation between each number and the number, carried out to all significant digits, that each was rounded from.

The front DoF and rear DoF displayed by both DoF Master and Cambridge are rounded numbers. Adding those two rounded numbers together can increase the variation between the rounded number and the actual solution for total DoF carried out to all significant digits. The total DoF for Cambridge is a number rounded after the more precise result of internal calculations using numbers extended to more significant digits is calculated. The total DoF for DoF Master is apparently the sum for the two rounded values for front and rear DoF, if DoF Master always displays a total DoF that is the exact sum of the two rounded numbers displayed to show front and rear DoF.

Claiming a number rounded at the last stage of a calculation is wrong because it is not equal to the sum of two numbers which were rounded before they were added together demonstrates a total lack of understanding of which number is actually more accurate.

 

[Michael Clark]

The highest DR 35mm sensors at DxO and PtP are high pixel density sensors. At DxO the top 9 spots are held by high density sensors.



This has not been my experience at 50mp, and I don't hear 90D or M6 mark II owners complaining at even higher pixel density.



And for a few stops beyond that you'll still see IQ gains.

Then one must ask what DxO's methodology for testing DR is and what they are actually measuring. (Hint: it's super top secret).

If DR is tested using scenes that have uniform brightness, then sensel size shouldn't matter, because the total number of photons increases at the same rate as the total area of the sample size. A sensor with photosites four time as large (2X wider and 2X taller) will have four times the full well capacity and would also collect four time as many photons if the light is uniformly distributed.

The problem with light is that it demonstrates a property know as Poisson distribution in which the exact distribution of photons within a light field is random. As the intensity of the light (thus the sample size per sensel) increases, the distribution becomes more uniform due to the laws of probability: as the size of the sample is increased the cumulative results of randomness will be more uniform. As the intensity of the light (the sample size per sensel) decreases, the distribution becomes more random. It's still the case, though, that if a scene being measured is illuminated very evenly, then the number of photons falling on each sensel will be more uniform than if a scene contains very dark areas with bright point sources of light in the middle of those dark areas. Something like the night sky, for instance. Or a test bench that projects microscopic points of light of varying intensity through a lens to a sensor.

Let's then say that one sensor does more on-die analog noise reduction than another before the signal is converted to binary numbers by the ADC. If we are measuring DR using point sources of light surrounded by very dark areas and the sensor with more aggressive analog NR eliminates the weakest point sources of light (such as the dimmest stars or the lowest intensity points of light projected on our test bench) in the field of view along with Poisson distribution noise, then that sensor will score better when using certain methods of measuring noise and thus measuring DR (where DR is defined as the difference between the noise floor and full saturation). This doesn't necessarily mean it actually has higher DR than another sensor that does not eliminate as much Poisson distribution noise prior to the ADC, it just means it's a bigger "star eater" which makes it look like it has a lower noise floor that it actually has. Will this result in cleaner images when one pushes the shadows? Absolutely it will. Does that mean the result more accurately depicts the actual scene that was photographed? No it does not. Quite the contrary. Does the sensor that produces the artificially "cleaner" images actually have higher DR than if it left the dimmest details in the scene along with more Poisson distribution noise? Nope.

 

[dtaylor]

you do it at macro (1:1) distances and see what you get.

Sorry it's not an AF chart. It's a bottle of Listerine at 45 degrees.

 

[dtaylor]

That's exactly what you wrote.

You claimed Cambridge's DoF calculator is inaccurate because a number rounded from the end result of an equation that used numbers extended to many more significant digits during all of the calculations does not agree with the sum of two numbers that were rounded before being added together.

It wasn't off by a small amount which would be the case with the type of rounding you describe. I wouldn't have mentioned it if it had been. It was off by a much larger factor but only in specific test cases. Now add the cases of extreme print sizes (small or large) where the calculator gives clearly wrong answers across all three fields. That's what leads to my suspicion of a rounding error in the formula, not in the final presentation.

 

[dtaylor]

Then one must ask what DxO's methodology for testing DR is and what they are actually measuring. (Hint: it's super top secret).

DXOMARK camera sensor testing protocol and scores

For our DxOMark camera sensor reviews, we measure the image quality performance only of camera sensors that are capable of capturing images in RAW format, and we do this before demosaicing or any JPG processing has taken place. You can read more about the DxOMark approach to image quality...

www.dxomark.com

You can also quite clearly see, in DPReview's DR comparison tool, that the D850 (higher pixel density) yields better results under a +5 or +6 push than pretty much anything else, even when starting at ISO 100.

As for full well capacity: as I recall it was in one of the threads of this forum that someone pointed out electronic noise does not scale in a linear fashion. So while a larger pixel has greater FWC, a smaller pixel can have a wider total range with lower noise at the floor.

 

[dtaylor]

you do it at macro (1:1) distances and see what you get.

Here's a 2nd example. I don't believe the first one actually started at 1:1. I'm adapting a Canon FL 50mm f/3.5 macro and I have to use the right adapters/tubes to get 1:1. (Lens is natively 2:1.)

All samples, first post and this one, were shot wide open. Mouthwash bottle at a roughly 45 degree angle. I racked out the lens to 1:1 and adjusted camera distance to get the b into focus. Then I moved back 1.6x, focused on the b using the lens, cropped to the 1.6x view and enlarged. Screen shot of the relevant section 2-up in PS.

 

[Michael Clark]

DXOMARK camera sensor testing protocol and scores

For our DxOMark camera sensor reviews, we measure the image quality performance only of camera sensors that are capable of capturing images in RAW format, and we do this before demosaicing or any JPG processing has taken place. You can read more about the DxOMark approach to image quality...

www.dxomark.com

You can also quite clearly see, in DPReview's DR comparison tool, that the D850 (higher pixel density) yields better results under a +5 or +6 push than pretty much anything else, even when starting at ISO 100.

As for full well capacity: as I recall it was in one of the threads of this forum that someone pointed out electronic noise does not scale in a linear fashion. So while a larger pixel has greater FWC, a smaller pixel can have a wider total range with lower noise at the floor.

I've read that plenty of times. It does not reveal DxO Mark's actual testing methodology for determining the noise floor that informs how they say, "This sensor has 13.7 EV dynamic range as tested and 14.2 EV when normalized to an 8 MP size (or whatever size it is to which they normalize). Just like it does not reveal, for instance, how different factors are weighted to arrive at a sensor's overall single number "Score".

Here's a 2nd example. I don't believe the first one actually started at 1:1. I'm adapting a Canon FL 50mm f/3.5 macro and I have to use the right adapters/tubes to get 1:1. (Lens is natively 2:1.)

All samples, first post and this one, were shot wide open. Mouthwash bottle at a roughly 45 degree angle. I racked out the lens to 1:1 and adjusted camera distance to get the b into focus. Then I moved back 1.6x, focused on the b using the lens, cropped to the 1.6x view and enlarged. Screen shot of the relevant section 2-up in PS.

View attachment 196712

When you say you moved back 1.6X, from where on your camera/lens did you measure your original distance to the "b" and from where did you measure that distance times a factor of 1.6?

Not to mention, as anyone can clearly see, the DoF of the second image is narrower than the first as the edges have much less contrast on the second image than they do in the first. Though in all fairness, it looks like the second image has other factors making the entire field blurrier as there is less overall contrast, even in the center of the field. Camera motion/shutter shock, perhaps?