[clackers]

The question is to know when it is worth it and if you also accept the downsides: more expensive, bigger/heavier gear.

(Ahem). I understand, Nicolas - I shoot both FF and APS-C.


https://www.pentaxforums.com/forums/26-mini-challenges-games-photo-stories/31...ml#post3467798

[Fries]

I can't wait for the new website update so we can start a new thread, which should be at least briefly uncontaminated by equivalence measurbation.

I don't expect 'them' to agree anytime soon, but I agree.

[reh321]

In my experience, the two things that most photographers would do well to learn are (a) to use a tripod more and (b) to use an external flash. Narrow depth of field does work for some photos, but I have had plenty of photos on APS-C spoiled because of too narrow depth of field -- making that even more shallow could be problematic.

Yes, at another photography discussion site where I'm a regular, at least once a month someone will post an image with focus issues and several people will immediately start talking about front/back focus; once the OP has answered a couple of questions, it turns out that the real issue is shallow DoF compounding a poor choice of focus point(s).

---------- Post added 01-03-16 at 07:28 PM ----------

I can't wait for the new website update so we can start a new thread, which should be at least briefly uncontaminated by equivalence measurbation.

Regardless of how many new threads are started on the subject, a sub-thread like this will live forever, and every time you think it has finally died, someone will feel a compulsion to get it going again.

[BrianR]

Tsk, tsk.


It's not true, Brian.


The K-3 has more pixels than an A7IIS or a Canon 1DX.

For pity's sake. You are far too intelligent to be so disingenuous or to quote completely out of context like this.

"The d810 and the k5 have the same size pixels (or close enough). The d810 has a larger sensor. Therefore, the d810 has more pixels, and due to downsampling will 'win' with regards to noise when it comes time to compare resulting images at the same print size. Bigger sensor => more pixels => less noise for a given print size."

In case it wasn't clear that the implications above were under the assumption of the same pixel size of the k5 and d810 comparison, my very next sentence was:

"Yes, it is true that a larger sensor does not guarantee more pixels, everyone knows this."

You're going out of your way to misunderstand and pick out fragments to argue with. You're better than this.

If you bothered to read the rest of my post you'd see the A7IIS and Canon 1DX are in no way a contradiction to anything I've said. They have less pixels than a k3, but the sensors have more surface area. Hence, they have ginormous pixels by comparison and you get a 'win' in this column with regard to noise. Enough of a win to make up for less down sampling with the 'print' comparison (of course there will be other factors involved with this real world comparison).

You're certainly one of the Good Blokes of the forum, but not for the first time you've mixed up causation with correlation

I like you too.

Sensor area is equal to the product of the number of pixels and the surface area of each pixel or do you disagree with this? (conceding that different manufacturing can result in different amounts of 'wasted' sensor space).

Increasing the sensor area necessarily has a positive impact on the combined effect of the number of pixels & pixel size* with respect to noise in a given print size or do you disagree with this?


* I repeat, one of these definitely goes up. One may go down, but then the other goes up even more to compensate.

[Fogel70]

Fogel, please prove that DxO Mark compensate for sensor size.

D810 and K5 score the same in screen comparison which means that D810 and K5 use same sized crop when comparing them.
So a 2.25x as large part of the captured image is used for K5 compared to D810. If you compare two different pixel sizes the camera with largest pixel will score the best regardless of sensor size.

FI the 12 year old 4MP APS-C camera Nikon D2H will show similar noise performance to Nikon D810 in screen comparison. https://www.pentaxforums.com/forums/16-pentax-news-rumors/310379-k1-pics-new-...ml#post3476449
Not because the images will show he same amount of noise, but because D2H use much larger pixels, so it will use much larger sized crop when compared to D810.
D2H has a sport score of 352 ISO and D810 has a sports score of 2853 ISO

In screen comparison the images is compared at 100% magnification at a fixed number of pixels so the smaller the pixels are the smaller crop will be compared. Why do you think the screen and print graphs show the same gap between the cameras if pixel count is the same when comparing two different sized sensors?

The 'Print' graph does no such thing.


For the last time, it's the result of *downsampling* (you can get Gimbal to explain it to you - he reckons he understands it!)

Yeah sure the lower noise in images from larger sensor is result of downsampling, but what if you have a APS-C sensor with more pixels that FF, would you believe that the APS-C sensor get best noise performance then as the image from the APS-C camera is downsampled more? Or what if the FF and APS-C sensor contain the same amount of pixels so they are downsampled the same amount?

The image captured by the larger sensor is magnified less when printed/viewed at a same size, so image noise is magnified less for a larger sensor. But the lower magnification of the larger sensor contain the same amount of image data, so signal to noise ratio improves for larger sensors.

The 36x24mm image data captured by the FF sensor in D810 has better signal to noise ratio than the 24x16mm image data captured by the APS-C sensor in K5 using same exposure settings. Each pixel on both of them captures the same signal to noise ratio.

The screen graph is the only genuine one showing noise captured in the camera image.

The print graph is a repeat analysis after noise has been removed by software algorithm.

The screen graph show pixel noise, and print graph show image noise. You do not seems to realize that those two are different. Where the first show noise in a varied sized crop that depend on pixels size, and the second use the whole sensor area to mimic what we experience when viewing/printing images.

Have you viewed imaged from K3 and D810 at the same size at same high ISO to see which i noisier? Or done the same thing with images from D750 and K3 II?
Do you understand the difference between pixel noise and image noise?

[clackers]

The screen graph show pixel noise, and print graph show image noise.

In your head only, Fogel, not according to DXOMark.


Nothing is cropped, and the print graph is no more 'the image' used than the screen graph - please stop saying that.


The print graph does not even involve a real world test - it's imaginary.


They do what they call 'normalize', and this is not what you think it is.


It is simply applying a theoretical formula to the only real test - the screen one - for what improvements 'should' occur. An extra column in the Screen Graph Excel worksheet, if you will:


Detailed computation of DxOMark Sensor normalization - DxOMark


If you shoot a picture of an 18% card with both the full frame and in cropped mode, you will *not* get the predicted 0.5eV increase in dynamic range. You will get close to zero.


Don't you love companies inventing metrics made to be misunderstood?

[clackers]

Increasing the sensor area necessarily has a positive impact on the combined effect of the number of pixels & pixel size* with respect to noise in a given print size or do you disagree with this?

If you say "likely" instead of "necessarily", you've nailed it for me, Brian.


There's no rule about what a manufacturer does - Sony do it two different ways with the A7S and A7R. A designer may want gaps between pixels instead of putting them so close together there's interference and diffraction. Canon deliberately went for more resolution than ISO performance in the FF 5DS with a 4.14 pixel pitch - smaller than the 7D!


I think it's best to do without the 'middle man', and talk only about those pixels. A discussion about are you better off getting a clean image to start off with or do you want lots of pixels to attempt to clean up a noisy image afterwards is for another day.


If you want to take anyone to task in - what did you call it - Breakdancing Thunderdome?, it's Gimbal who says "Total sensor size have a large impact on noise, pixel size not so much, some impact but not nearly as much as the total sensor size."

[Fogel70]

In your head only, Fogel, not according to DXOMark.


Nothing is cropped, and the print graph is no more 'the image' used than the screen graph - please stop saying that.

Then how do you explain that Nikon D2H ger similar score in screen comparison as Nikon D810. Please try to answer my questions for once, even the once I asked in previous post!
Do you still agree that in screen comparison you compare FI a APS-C crop of D810 with the full sensor of K5, as they use the same sized pixels?

Try to prove what you say is true instead of just repeating same sentence over and over again.

It is simply applying a theoretical formula to the only real test - the screen one - for what improvements 'should' occur. An extra column in the Screen Graph Excel worksheet, if you will:

Yes it is an theoretical calculation that is base on what happen when you view or print the images at the same size.
If you compare the result with practical tests you will find them being accurate.

If you shoot a picture of an 18% card with both the full frame and in cropped mode, you will *not* get a 0.5eV increase in dynamic range. You will get zero.

Where do you get that from?

[Nicolas06]

(Ahem). I understand, Nicolas - I shoot both FF and APS-C.
https://www.pentaxforums.com/forums/26-mini-challenges-games-photo-stories/31...ml#post3467798

In your head only, Fogel, not according to DXOMark.

Nothing is cropped, and the print graph is no more 'the image' used than the screen graph - please stop saying that. The print graph does not even involve a real world test - it's imaginary.

Clackers you troll here and try to confuse people. You admited basically that FF are better in practice but you continue to try to confuse people with print and 100% crops. This is counterproductive.

The print results are here to match what you would get doing a high quality 300dpi A4 (approx 8"x12") print. That mean actually you don't need more than 8MP and either your editing software or the printer driver will downsample to 300dpi so 8MP. And the results DxO shows actually match the real world results.

All the rest you discuss here is more confusion than anything else.

[normhead]

So how do you suggest ISO noise test should be done and presented to emulate having to stop down an extra stop on FF compared to APS-C?
I mean, stopping down the lens one extra stop on FF do not change anything in the measurement as one stop slower shutter speed will then be used.
When shooting test charts you do not need much DOF.

Sorry, but that's the 36x24 cool aid speaking. There have always been and will always be instances where shutter speed is important.

I mean, stopping down the lens one extra stop on FF do not change anything in the measurement as one stop slower shutter speed will then be used.

Classic jshermanism. That is the classic propagandists argument. The one stop stopping down is irrelevant, because ____<fill in the blank> so the 36x24 has a noise. advantage.

Is that a fact? Really?


So I could have kept my DoF and had the same image here


or here


This is from yesterday's shooting, I kept them out of interest, but bottom line, more shutter speed equals more images like this


And fewer like those.

How do you win the 36x24 noise argument? You claim that one of the metrics is irrelevant. Depth of Field is irrelevant, shutter speed is irrelevant, something always doesn't count. With all due respect, this is like little kids up on the playground making up rules to some game they are playing, so they win.

The fact is, sometimes shutter speed is relevant, sometimes Depth of Field is relevant. Sometimes (as in the above images) they are both relevant at the same time.Now there are photographers for whom depth of field is kind of a fuzzy thing. They've never been trained to pay attention to it. In fact they've never had a camera where you can actually see it in their little viewfinder. So for the untrained photographer it is easy to make that kind of statement. For those of us who have been in the studio struggling to get all of a product in focus with an 8x10 or 4x5 film camera, we are acutely aware of depth of field , and for those of us who understand that when we are struggling with noise, we are always also struggling with reciprocity, we are also acutely aware of what is happening with our light intensity and shutter speed.

Apparently many of these people who profess 36x24 noise advantage are not fully understanding those things, or they wouldn't be making this same mistake over and over again.

Same picture same noise. You cannot change this without altering the picture. That's equivalence. "It doesn't matter if I change the picture or this or that function." isn't an argument. There is always an instance where it does, that negates that type of argument, every one of them.

[BrianR]

If you say "likely" instead of "necessarily", you've nailed it for me, Brian.

If "necessarily" is false, you should be able to come up with specs for a FF camera with a combination of "pixel size" and "pixel number" that hurts its noise performance relative to a k3 (when images are viewed at the same size). Something within the realm of possibility. i.e. no taking the k3's 24mp, keeping the same sized pixels but spreading them all around the larger area with huge gaps in between. No taking those 24mp and shunting them to the outer edges because you've always wanted film shaped like a doughnut.

A designer may want gaps between pixels instead of putting them so close together there's interference and diffraction.

Of course, changes in manufacturing can result in better performance. "Assuming roughly equal technology" should probably be stated at every turn- I think you'll find most don't expect a FF with the technology from 10 years back to stand up as well with an aps-c from today as it did 10 years ago.

It's my (possibly naive) understanding that for sensors from a similar time period things like gaps between pixels are pretty minor differences relative to pixel count and pitch.

There's no rule about what a manufacturer does - Sony do it two different ways with the A7S and A7R.

Canon deliberately went for more resolution than ISO performance in the FF 5DS with a 4.14 pixel pitch - smaller than the 7D!

And in both pairings, the predictable happens. The smaller pixels of the A7R make it noisier than the A7S, until you down sample to the same viewing size than it's a wash. Similar for the Canons, the 5DS is as noisy (pretty close) as the 7d, then the 7d is crushed after down sampling.

I think it's best to do without the 'middle man', and talk only about those pixels. A discussion about are you better off getting a clean image to start off with or do you want lots of pixels to attempt to clean up a noisy image afterwards is for another day.

The middle man is unavoidable. In both the practical (how the k5 vs k3 perform at a given print size) and the theoretical (product of pixel count and pixel pitch necessarily changes with sensor size).

If you want to take anyone to task in - what did you call it - Breakdancing Thunderdome?, it's Gimbal who says "Total sensor size have a large impact on noise, pixel size not so much, some impact but not nearly as much as the total sensor size."

Sure, I'll give it a crack (he's invited to correct my understanding of his words). Cue Tina Turner singing Michael Jackson's "Bad" while Mel Gibson moonwalks around Blaster.

"Total sensor size have a large impact on noise"

Accepting down sampling to a common viewing size, sure.

"pixel size not so much, some impact but not nearly as much as the total sensor size."

In a comparison between sensors of the same size (and roughly the same date of manufacture:P), and coupled with down sampling, also sure. Many real world examples (k5 vs k3 being the most relevant to Pentax Forums) support that changing the pixel size under these constraints has little effect on noise in your final image. Evidence to support the conclusion it has some impact at a given print size is probably visible in real comparisons (i.e. k3 vs k5), but I'd suggest any observable differences here are too small to differentiate from the necessary changes in manufacturing (which I'm assuming are also pretty 'small', but of a similar order of magnitude to the observed differences between the k3 and k5), tough to isolate, and also not backed up by theoretical models.

[beholder3]

Same picture same noise. You cannot change this without altering the picture. That's equivalence.

Yup.

Actually I could be evil and point out that all these noise discussions tend to focus on the one extreme end, with little light, where you open up the aperture a lot. So you use fast glass.

In reality (that now has nothing to do with the equivalence rule of thumbs) you will then find that large image circle fast lenses suffer a lot more from vignetting. Vignetting actually needs to be countered by what?
Right, pulling up the shadows. And that does what?
Right, increase the noise.
Now we have all those MTF chart fanatics who claim that corner resolution is key, because the corners are important (remember DA 18-135?; everybody loves to put the main subject in the corners). So obviously noise performance in the corners is key as well, isn't it?

Now to ground us all in reality we look at 24mm on APSC versus 35mm on FF: Vignetting = Noise performance:
Canon EF 24mm f/1.4L II USM Lens Vignetting
That is -1 EV compared to -3 EV, so a whopping 2 full stops worse noise performance (in the outer image regions) for the identical picture on FF with top notch lenses.

And remember: The pictures would otherwise had the identical noise performance if created comparable.

Is this being raised a lot by the proponents in discussions? Why?
Fast short lenses are not relevant? Shooting open is not relevant? Corner performance is not relevant?

I think FF has it's place in the world. But please stop the nonsense biased arguments.

[Fogel70]

Sorry, but that's the 36x24 cool aid speaking. There have always been and will always be instances where shutter speed is important

Yes, but what I was referring to was when testing cameras. Then larger sensors will score better at every ISO, unless the camera use old obsolete sensor. Or do you suggest testers start underexpose images one stop when testing noise on FF camera, so the result comply with equivalence to APS-C?

In practical use there might not be much difference as you may have to use higher ISO on larger sensors cameras. Some say 1" sensor is the ultimate sensor size, some say mft, some say APS-C, some say FF.... And they may all be right.

[normhead]

Yes, but what I was referring to was when testing cameras. Then larger sensors will score better at every ISO, unless the camera use old obsolete sensor. Or do you suggest testers start underexpose images one stop when testing noise on FF camera, so the result comply with equivalence to APS-C?

In practical use there might not be much difference as you may have to use higher ISO on larger sensors cameras. Some say 1" sensor is the ultimate sensor size, some say mft, some say APS-C, some say FF.... And they may all be right.

Too, funny, everyone does it so it must be right. Having had academic parents, that kind of argument would have seen me laughed into the next county. That's not proof. That's an admission that you're a victim of herd mentality.

Resist the herd.

What does this have to do with under-exposing? If you understand equivalence, you'd understand under-exposing is not part of the equation.

APS-c- ISO 100, ƒ5.6, 1/50s
36x34- ISO 200, ƒ8, 1/50s

No image is under-exposed, both have the same DoF, the image is made with the same total light, since you have the same signal, you have the same signal to noise. They are functionally the same image with a few caveats.

It's not rocket science.

I can't explain this any simpler.

The big question I would like answered, is why did all those people on all those sites publish all those tables, comparing different images?

The question is why did all those folks you talk about do this?
APS-c ISO 100, ƒ5.6, 1/50s
36x24 ISO 100, ƒ5.6, 1/50s

The 36x24 image uses double the light and therefore has a stronger signal and has better signal to noise... but it's not the same picture, the DoF is different.

So you want to know why they did this? They are all living in the "36x24 is always better world", and they were looking for proof of their propaganda, not trying to provide useful information. But that was probably just the first guy who proposed the argument (or maybe not even him). The others didn't know enough about what they were talking about and just parroted what the guy they thought was the smart guy said. And there are a lot of people on this site who got sucked in. Many that are still taken in by it.

You could do post graduate work using this as an example of

The funny thing about this is, falconeye, the guys on luminous landscape, everyone who really does understand the physics have always said this was true.
They always said, equivalence didn't mean, one system was better than another. They just didn't waste their time breaking it down so everyone could understand it.

I'm not saying there is not any advantage to 36x24, at the bottom of the spectrum, shooting a lens wide open on both, there is a Narrow DoF, 1 stop better signal to noise image, you can take on a 36x24 camera, you can not duplicate on APS-c. There are bokeh effects in the 35 to 50mm range, you cannot duplicate on APS-c. There are image distortion issues in UWA that may or may not be more pronounced on APS-c. when shooting say 15mm 36x24 and 10mm APS-c. And if you wish to increase your resolution to 36 MP or beyond, you have to go larger format. You do get something for your money. But, it's not as much as many people will tell you. In 90% of the range of possible images, you can take the same image on both systems, there just is not that much difference between them. APS-c and 36x24 are much more the same, than they are different.

And people have said this over and over.

Out of personal preference, there's absolutely nothing wrong with wanting to have that extra 10% for double or triple the price. And a pro may feel obligated to make that extra 10% available to his or her clients, well, because he's pro, but many don't feel any such obligation. Just keep it all in perspective. But don't delude yourself into thinking that by simply going to 36x24 you will make every image better. In fact the opposite is true, If you can take the image with APS-c, odds are, it will be the same, especially for similar costing systems of the same MP.

[Fogel70]

Too, funny, everyone does it so it must be right. Having had academic parents, that kind of argument would have seen me laughed into the next county. That's not proof. That's an admission that you're a victim of herd mentality.

What does this have to do with under-exposing? If you understand equivalence, you'd understand under-exposing is not part of the equation.

APS-c- ISO 100, ƒ5.6, 1/50s
36x34- ISO 200, ƒ8, 1/50s

The underexposing is so you can measure the same amount of noise at the same ISO on both APS-C and FF, because it seems to be breaking some sort of taboo suggesting that FF has better ISO noise performance than smaller sensors.

When I earlier suggested "The normal procedure to compare noise is to compare at the same ISO with same light conditions." it was commented as "You mean the endless pro full frame drivel that gets posted on the web?..." and "Then your "normal" is apples versus oranges..."

And when I suggested "I mean, stopping down the lens one extra stop on FF do not change anything in the measurement as one stop slower shutter speed will then be used" it was commented as "Sorry, but that's the 36x24 cool aid speaking..."

So by using...
APS-c- ISO 100, ƒ5.6, 1/50s
36x34- ISO 100, ƒ8, 1/50s
Both capture the same same image and same ISO noise performance can be measured on both for the same ISO.
Everyone happy?

The big question I would like answered, is why did all those people on all those sites publish all those tables, comparing different images?

The question is why did all those folks you talk about do this?
APS-c ISO 100, ƒ5.6, 1/50s
36x24 ISO 100, ƒ5.6, 1/50s

The 36x24 image uses double the light and there for has a stronger signal and has better signal to noise... but it's not the same picture, the DoF is different.

They use these exposure settings because they measure noise at a specific ISO, using equivalent DOF would not change anything as they then have to compensate with shutter speed to not underexpose the FF image, and the larger sensor will still capture twice as much light and the ISO noise performance would be exactly the same as if both use the same f-stop.