My comparison of sensor size and print quality

[woof woof]

Unfortunately, I don't think it's as simple as that with digital :-(

I suppose with film the age of the camera had next to no effect assuming the seals weren't gone and we're using the same film and processing and the film is held held reasonably flat in the camera etc. With digital time moves on and if we're lucky newer generations of kit creep ahead of older kit and even of older larger format kit. For example when I first bought into early MFT I was convinced that in some situations it gave better image quality than my FF Canon 5D.

Usually, unless the manufacturer is behind the curve and that can happen, usually a bigger format of anything like a similar tech generation to small format kit will provide better image quality. I'm ignoring lenses here and just talking about file quality.

Just on my own experiences of comparing sensor sizes in print. I have a Medion compact camera which is frankly awful by just about any standard and yet I can do a print which will look very nice when filling an A4 sheet. However, just as I can pixel peep on my screen if I look very closely at a Medion print the differences between it and a print from FF Sony are there to be seen.

Processing software definitely has an effect on IQ too and when I've seen a significant improvement I've reprocessed old raws. Being able to revisit old files as processing technology improves is quite a big plus point for shooting raw.

 

[sk66]

1. Raw black/white point
2.Demosaic
3. Input color profile (seems to convert the original device colour space into a working colour space)
4 output colour profile (seems to convert between colour spaces)
5. display encoding

1. Discards noise (black point) and clipped values (white point).
2. This is technically referred to as "interpolation" (determining R/G/B values for each pixel). There is no color space here... the numbers have no real meaning.
3. Input color profile is things like white balance, and device color specifics. And some of those device color specifics are set in the camera profile i.e. how they want the colors rendered and other choices (e.g. a user settable white balance/hue correction). This step is creating/defining the standard device (camera's) color space. Changing this requires the raw data because the individual R/G/B values may need changed for every pixel, and it is not linear/uniform. The output from this is a "standard" color space and now linear.
4. Output color profile is the working (viewing) color space (e.g. Melissa in LR). This step forward does not require individual pixel raw data values, and can be applied to any file type.

Some references do make it seem like interpolation and demosaicing are the same thing, but they are not. Interpolation is just the math, demosaicing is the math + color science (e.g. human vision).

 

[sk66]

A fascinating read, thank you @myotis.

Just checking if my simple understanding is correct - before processing, GFX files look better than Nikon files. With some processing effort, Zf prints could be matched to the GFX more easily than Z8 prints could (despite the Z8's higher resolution). But we don't have a comparison of processed GFX vs processed Nikon.

This would generally be the case with any MF/FF comparison; much like FF/APS.

I.e. some prefer the Hasselblad X1D 50c which has the same sensor. But it also has a 16 bit ADC and a different color space/rendition (Hasselblad Natural Color).

 

[myotis]

Interesting. I see a lot of people raving about GFX images but usually only when they've zoomed in to a single eyelash. And frankly, who cares? But this is a practical test.


I wonder - have you tried any form of blind randomisation? eg print comparable images, but then have someone else label them & shuffle them before you get to look? That would be illuminating.

No, but...

I made the prints, labelled them on the back, and then took them into a room with better lighting and looked at them without knowing which was which.

I then asked my wife (nearly 20 years as a professional photographer, a long time ago) to do the same,

In terms of matching the Nikon prints to the Fuji prints, we repeated this viewing exercise with me tweaking the prints from the Nikon cameras after each comparison, until we confused which print came from the Fuji and which one came from a Nikon. At that stage, even though some minor differences remained, we decided they were close enough to be called "matched"

 

[myotis]

1. Discards noise (black point) and clipped values (white point).
2. This is technically referred to as "interpolation" (determining R/G/B values for each pixel). There is no color space here... the numbers have no real meaning.
3. Input color profile is things like white balance, and device color specifics. And some of those device color specifics are set in the camera profile i.e. how they want the colors rendered and other choices (e.g. a user settable white balance/hue correction). This step is creating/defining the standard device (camera's) color space. Changing this requires the raw data because the individual R/G/B values may need changed for every pixel, and it is not linear/uniform. The output from this is a "standard" color space and now linear.
4. Output color profile is the working (viewing) color space (e.g. Melissa in LR). This step forward does not require individual pixel raw data values, and can be applied to any file type.

Some references do make it seem like interpolation and demosaicing are the same thing, but they are not. Interpolation is just the math, demosaicing is the math + color science (e.g. human vision).

I think we are getting closer

Like Darktable I think of interpolation and demosaicing as the same thing, with the Input color profile being applied to the RGB values output from the demosaicing.

Where are you getting the information that the Input profile is changing the interpolated values rather than just acting on the already interpolated ones?

 

[sk66]

Where are you getting the information that the Input profile is changing the interpolated values rather than just acting on the already interpolated ones?

What is the difference here? The RGB values per pixel are different in the camera color space prior to conversion into display color space. I mean, the interpolated values at that stage don't change, but they also have no real meaning as such and they are not in linear RGB. They must be "changed" before they are of use visually.

This is from the RawTherapee documentation:

Input Profile​

An essential first step in raw processing is the faithful conversion of the camera sensor data to an internal RGB color space. This conversion requires an input profile made specifically for the camera. Such a profile is the result of the analysis of how specific colors and tones are captured, processed and represented as raw data by the camera (for more details, see e.g. Elle Stone's article, DCamProf's documentation or How to create DCP color profiles. Without a camera-specific input profile, accurate color representation is impossible.

I.e. the same RGB values represent very different colors in different color spaces and the first color space of importance is the camera's (here there can be some confusion/mixing between referring to the camera's un-demosaiced colors and input corrected colors). And color space changes are not linear/proportional RGB changes so they can't just be applied like a layer/filter. And since it is the first color space in the chain, it has to go back to the raw data.

Here's an example of the same colors in sRGB and Adobe RGB.

 

[myotis]

What is the difference here? The RGB values per pixel are different in the camera color space prior to conversion into display color space. I mean, the interpolated values at that stage don't change, but they also have no real meaning as such and they are not in linear RGB. They must be "changed" before they are of use visually.

Because this is the only question I have been discussing :-(

I know you have been discussing all sorts of other things, despite me trying to bring it back to the original question: does the camera profile alter how the camera raw data is interpolated, or is it applied after the camera raw data has been interpolated.

ie at what stage in the raw processing pipeline is the camera profile applied.

Now, it's always possible that different raw processor handle this differently (e.g. scene referered vs screen referred), but Google's AI summary says the profile is applied after the interpolation (but I don't trust Googles AI) . Questions answered on the Adobe forum and the Darktable forum also say it's applied after the interpolation, but I have no idea who the people answering the questions are.

So while it isn't actually all that important, after all this time, I still feel the need to find an answer.

This is from the RawTherapee documentation:


I.e. the same RGB values represent very different colors in different color spaces and the first color space of importance is the camera's (here there can be some confusion/mixing between referring to the camera's un-demosaiced colors and input corrected colors). And color space changes are not linear/proportional RGB changes so they can't just be applied like a layer/filter. And since it is the first color space in the chain, it has to go back to the raw data.

Here's an example of the same colors in sRGB and Adobe RGB.
View attachment 489512

I'm not sure this tells you anything about when the profile is applied.

It depends on what is meant by the raw data. is it the camera raw data or the interpolated raw data? Or are the interpolated raw data values no longer considered "!raw" data.

 

[sk66]

Because this is the only question I have been discussing :-(

As I originally stated, I've always understood demosaicing to be the process of generating a viewable image with accurate/useful colors (as intended by manufacturer). Because saying "demosaicing creates a viewable/editable image" is wrong... An un-demosaiced image is viewable/editable; it's just not of much use, and an interpolated image isn't really much better.

But I'll have to admit I was wrong in that demosaicing and interpolation are more the same thing than I believed/understood; the color science applied stops at human vision modeling and does not include target (camera) color space. So my use of "demosaicing" to mean making an "accurate/useful" viewable image was incorrect. It is rare for me to be so adamantly wrong, but was/am (in my defense, I was aided with a fair bit of AI confirmation bias when self checking).

 

[myotis]

As I originally stated, I've always understood demosaicing to be the process of generating a viewable image with accurate/useful colors (as intended by manufacturer).

And that was when I responded to say that I saw the demosaicing as a distinct process within the raw processing pipeline, and thought the camera profile was applied after the interpolation step, (need to turn the interpolated data into a viewable/editable image) rather than it being used to directly affect the interpolation.

I thought we then both set out to find out which view was correct.

Because saying "demosaicing creates a viewable/editable image" is wrong... An un-demosaiced image is viewable/editable; it's just not of much use, and an interpolated image isn't really much better.

But I'll have to admit I was wrong in that demosaicing and interpolation are more the same thing than I believed/understood; the color science applied stops at human vision modeling and does not include target (camera) color space. So my use of "demosaicing" to mean making an "accurate/useful" viewable image was incorrect. It is rare for me to be so adamantly wrong, but was/am (in my defense, I was aided with a fair bit of AI confirmation bias when self checking).

AI (artificial idiocy) is a pox on modern society. I watched a documentary a few months ago that was reviewing AI assistants, and they were all programmed to give answers they thought the questioner would want to hear as well as always be positive and encouraging regardless of the situation.

Rant on:

AI is a great discovery with tremendous potential for good, but it's going to destroy the human race well before climate change does. Having said that, our increasing addiction to UPFs might get there before either of them !

Rant off.

 

[juggler]

And that was when I responded to say that I saw the demosaicing as a distinct process within the raw processing pipeline, and thought the camera profile was applied after the interpolation step, (need to turn the interpolated data into a viewable/editable image) rather than it being used to directly affect the interpolation.

Danger.. there's a significant rabbit hole awaiting the unwary

It's a long time since I looked at this in any detail and the technology may have moved on - and my point of reference is industrial vision systems.

Anyway: there are many different demosaicing & interpolation algorithms. It's more useful to think of it as a series of interpolation operations in a pipeline; they can insert camera profiles, median filtering, clipping, contrast curves and white balance adjustments in a variety of different sequences.

Here's one example: https://docs.baslerweb.com/visualap...ntent/examples bayer white balancing 5x5.html

 

[TimHughes]

... and I imagine the actual colours used in the mosaic filters vary between manufacturers and maybe models, adding another important factor to the final results.

 

[thedistrictline]

Swinging this back around, could you try an asp-c file against the z8 or zf, itll help me reduce my GAS

 

[sk66]

... and I imagine the actual colours used in the mosaic filters vary between manufacturers and maybe models, adding another important factor to the final results.

The spectral response curves vary between cameras for a multitude of reasons. And it's really a huge oversimplification when we talk about the CFA/photosites being RGB; they're really "long/medium/short" with overlaps. But cameras using the same sensor will have the same response curves even in cameras of different brands (the Mamiya/Hasselblad comparison).

A lot can also be attributed to the lens and default sharpening (clarity/contrast/etc) applied that you don't know about. I remember when we thought some lenses were getting "too sharp" for portraiture... i.e. what's the difference between lack of detail and smooth gradation?

 

[myotis]

Danger.. there's a significant rabbit hole awaiting the unwary

It's a long time since I looked at this in any detail and the technology may have moved on - and my point of reference is industrial vision systems.

Anyway: there are many different demosaicing & interpolation algorithms. It's more useful to think of it as a series of interpolation operations in a pipeline; they can insert camera profiles, median filtering, clipping, contrast curves and white balance adjustments in a variety of different sequences.

Here's one example: https://docs.baslerweb.com/visualapplets/files/manuals/content/examples bayer white balancing 5x5.html

Thanks but....

My post: "raw processing pipeline"

Your reply : it's more useful to think of it as a series of interpolation operations in a pipeline"

I know the banter between me and Steven has lasted many posts, but it has been mainly about the order of processes within the pipeline.

The Darktable information gives a bit of information on the different demosaicing approaches., but I agree its more complex than just a "demosaicing"

 

[myotis]

Swinging this back around, could you try an asp-c file against the z8 or zf, itll help me reduce my GAS

Actually, as an afterthought I did compare my 16mp Fuji X100s with the Zf, including buying Cobalt camera profiles for it.

Bear in mind that I have set a quality threshold at a 20" x 16" print level, and this is all very subjective.

The initial issue was that the Fuji lens came nowhere close to matching the quality of the Voigtlander Apo-lanthar on the Zf. But this was down to edge performace, and with Capture One, I can manually correct for loss of edge sharpness, which is what I did.

I only made a couple of prints for this comparison: at ISO 800 as it's higher iSOs that I am more likely to use with X100s.

I struggled to tell the difference between the finalised prints. Indeed in my final comparison print I over-sharpened a tiny wee bit (QImage sharpening set to 15 rather than 10) and the X100s print looked better than the Zf print.

On close inspection, you could see a bit of noise on the X100s print, which would have easily disappeared with some noise reduction, and it was far from objectionable in the subjects I was photographing.

My results have made me rethink my X100s as I was keen to replace it with a newer model, but now, I don't think I need to, at least not on an image quality basis.

However, it reinforces the theme of all my comparisons: the smaller the sensor, the more effort I seem to need to put into the processing.

Edit: I've just had another look, and if pushed I would say there was a small improvement in colour gradation with the Zf files.

 

[thedistrictline]

Great thanks! With my 26MP X-t30 (first gen) I'll take that as a success!

 

[Dave Canon]

There are colour differences between individual cameras of the same brand and model. I only use Raw and have individual conversion profiles for each of my camera to use in LR. A few years ago a member of my Camera Club (a professional portrait photographer) decided to do a similar exercise and produced prints from ten different cameras from a very small sensor point and shoot to the latest Nikon full frame. At the meeting of my Digital Imaging Group, we were all invited to indicate the order o quality from 1 to ten. There were few surprises almost everyone scored the same with the FF Nikon top and the small sensor camera bottom. In between they were also logical in respect of sensor size. The couple of variations were by people who accepted that they had less than perfect vision. In any case, when comparing image quality, dynamic range is also a key design criterion.

Dave

 

[myotis]

There are colour differences between individual cameras of the same brand and model. I only use Raw and have individual conversion profiles for each of my camera to use in LR. A few years ago a member of my Camera Club (a professional portrait photographer) decided to do a similar exercise and produced prints from ten different cameras from a very small sensor point and shoot to the latest Nikon full frame. At the meeting of my Digital Imaging Group, we were all invited to indicate the order o quality from 1 to ten. There were few surprises almost everyone scored the same with the FF Nikon top and the small sensor camera bottom. In between they were also logical in respect of sensor size. The couple of variations were by people who accepted that they had less than perfect vision. In any case, when comparing image quality, dynamic range is also a key design criterion.

Dave

This actually mirrors my results "until" I "matched" the prints, with, as I described, the smaller the sensor the greater amount of effort needed to achieve a match.

But, again, as described, I chose not to optimise the prints from the GFX. I just took the "default" GFX prints as the quality standard that the prints from the smaller sensors had to match.

Had I optimised the prints from each sensor and then made the comparison, I may well have come to a different conclusion. But the quality from the GFX, even at my default settings, made prints that were so much better than the smaller sensor's prints that for my purposes I was happy to work with the defaults (ie the Capture One and QImage One defaults)

The difference in colours between the cameras was an issue that initially confounded the comparisons, which is why I splashed out on the Cobalt camera profiles, which, as advertised, gave almost identical colour and contrast results for all four cameras I compared.

 

[myotis]

Great thanks! With my 26MP X-t30 (first gen) I'll take that as a success!

In this video, by Nigel Danson, he compares a print from the 45Mp Nikon Z8 and the 20mp Nikon Z50 II.

He can see differences in Lightroom, but on an A3 print he suggests he cannot see any differences, but also uses phrases like "absolutely similar" and "not a lot of difference".

With the A2 print he is now seeing differences, with the Z50II print slightly softer than the Z8 print, but he isn't sure if this is because of differences in lens quality, sensor size, or a combination of both.

He discusses print quality between 19 and 21 minutes into the video.

View: https://www.youtube.com/watch?v=a_6MyUnLGuU&t=2s

 

[thedistrictline]

Yeah I've watched this before , at 16x20 I'm not sure anyone would be able to tell at a normal viewing distance (cue rage bait), although we are the annoying people sticking our noses up to the wall

 

[sk66]

There are colour differences between individual cameras of the same brand and model.

Interesting. I've never heard/read of it before; but I've also never owned multiple bodies of the same model concurrently. Are the differences significant and based on a standardized/empirical test (e.g. calibrite color calibration)? I ask because I would be surprised if they were very significant/notable; but not that there would be "some differences" (there's always "tolerances").

or a combination of both.

Probably both, but it's probably not due to MP/pixel size. Some of the DX Z lenses are really quite inexpensive and there's probably some compromises being made. But moreso (IMO) this gets into the realm of "equivalence" for different sensor sizes; where the same camera settings *do not give the same results in terms of exposure (light gathered), depth of field, diffraction, etc, (with the smaller sensor suffering in comparison).

It also probably gets into optimal print sharpening requirements, which typically would be different for different amounts of enlargement (sensor sizes)... I say "probably" because I don't know what he did (nor what "should have" been done). But at A2 the Z50 is still providing 240 ppi, which is too small to see with the naked eye and any kind of typical images/viewing conditions (examination with a printer's loupe; shure).

In almost every case it comes down to total light (sensor size) and recorded resolution. With recorded resolution typically limited to something well below sensor resolution due to spherical aberrations, diffraction, aa filter, motion blur, etc (it is never equal to sensor resolution).
And something like leaves/small branches tend to add some random variability to the equation as well; as the shutter speed required to freeze motion is not the same between formats. And even if you do freeze the motion adequately, it may not account for a change in position/distance within the differing DOF's (not much of a factor at long distances).


(*everything records the same on a per area basis given the same composition. But the smaller sensor is enlarged to a greater degree which makes every defect/aspect more readily apparent/visible)

 

[myotis]

Interesting. I've never heard/read of it before; but I've also never owned multiple bodies of the same model concurrently. Are the differences significant and based on a standardized/empirical test (e.g. calibrite color calibration)? I ask because I would be surprised if they were very significant/notable; but not that there would be "some differences" (there's always "tolerances").


Probably both, but it's probably not due to MP/pixel size. Some of the DX Z lenses are really quite inexpensive and there's probably some compromises being made. But moreso (IMO) this gets into the realm of "equivalence" for different sensor sizes; where the same camera settings *do not give the same results in terms of exposure (light gathered), depth of field, diffraction, etc, (with the smaller sensor suffering in comparison).

In everyday photography, It's difficult to separate the two.

It also probably gets into optimal print sharpening requirements, which typically would be different for different amounts of enlargement (sensor sizes)... I say "probably" because I don't know what he did (nor what "should have" been done). But at A2 the Z50 is still providing 240 ppi, which is too small to see with the naked eye and any kind of typical images/viewing conditions (examination with a printer's loupe; shure).

I've used QImage to make prints for a couple of decades, bought on reputation, and certainly at the time, it made prints that looked so much better than i could make following the instructions on how to make prints with Photoshop?LR. Apart from an occasional quick print, I print everything through QImage and I am always pleased with the sharpening.

Between QImage optimising sharpness and based on 240ppi, sharpness wasn't really an issue with my test prints, until I got to the X100s, when problems with lens sharpness kicked in. So I wasn't pushing anything to the limits. But that was kind of the point of my tests.

In almost every case it comes down to total light (sensor size) and recorded resolution. With recorded resolution typically limited to something well below sensor resolution due to spherical aberrations, diffraction, aa filter, motion blur, etc (it is never equal to sensor resolution).
And something like leaves/small branches tend to add some random variability to the equation as well; as the shutter speed required to freeze motion is not the same between formats. And even if you do freeze the motion adequately, it may not account for a change in position/distance within the differing DOF's (not much of a factor at long distances).


(*everything records the same on a per area basis given the same composition. But the smaller sensor is enlarged to a greater degree which makes every defect/aspect more readily apparent/visible)

 

[Dave Canon]

Interesting. I've never heard/read of it before; but I've also never owned multiple bodies of the same model concurrently. Are the differences significant and based on a standardized/empirical test (e.g. calibrite color calibration)? I ask because I would be surprised if they were very significant/notable; but not that there would be "some differences" (there's always "tolerances").

A friend was a weekend wedding photographer and like to carry two identical FF Nikons but with different lenses but he found they produced different colours which is critical for weddings and fashion photography ( Martin Evening emphasises this in his books). I use the colour checker passport but use Adobe's DNG editor rather than the software which came with the passport. While precision colour was needed by my friend, I do not need such precision but it is still an ideal starting point even if I intend to be creative. It saves me a lot of time. The software can also create a profile for my infrared converted camera.

Dave

 

[sk66]

with different lenses but he found they produced different colours

That's not too uncommon. Different lens materials, combination/number of elements, coatings, and aberrations, can all alter the color rendition of one lens design compared to another.

 

[Dave Canon]

That's not too uncommon. Different lens materials, combination/number of elements, coatings, and aberrations, can all alter the color rendition of one lens design compared to another.

It was not due to the different lenses as he tried the same lenses on each camera and the colour differences were still present. He knew what he was doing as well as being a weekend wedding photographer his day job was a nuclear scientist.

Dave

 

[Ed Sutton]

his day job was a nuclear scientist.

I'd be more impressed if he was a rocket scientist.

View: https://www.youtube.com/watch?v=THNPmhBl-8I

 

[woof woof]

There's no doubt that different lenses give different looks but it may be possible to make the end result largely the same post capture. There's also no doubt that different cameras give different looks but again it may be possible to alter things post capture but all this fiddling costs time and potentially money too if someone is getting paid along the way. I suppose one way to get consistent results more easily is to go for a set of colour matched cini lenses and if you are using multiple cameras make sure they're all the same model of camera or at least cameras which gave the same colour responses.

On mp counts, one advantage of a higher mp count is that you stand a higher chance of being able to crop and still end up with a large image. This has obvious advantages including being able to use smaller lighter and cheaper kit and still being able to make a subject larger in the final frame. You can't always move closer to your subject and even if you can foot zoom alters perspective.

 

[myotis]

There's no doubt that different lenses give different looks but it may be possible to make the end result largely the same post capture. There's also no doubt that different cameras give different looks but again it may be possible to alter things post capture but all this fiddling costs time and potentially money too if someone is getting paid along the way.

To repeat one of the points I raised earlier. One of my initial problems in getting matching prints was that the Fuji GFX had very different colours to the Nikons, and although this came later, the Fuji X100s had very different colours to both the GFX and the Nikons.

I solved this (at some expense) by buying Cobalt camera profiles for each camera (used as the camera profile defaults for each camera). These worked as advertised and gave very similar starting colours for all four cameras.

I also mentioned (but never used) that Capture One has a tool designed to correct for differences in colour due to using different lenses. You photograph the subject through an opaque plastic filter, which C1 analyses,along with the actual subject file, to produce an "LCC" profile, which C1 then uses to correct any colour cast due to the lens.

I suppose one way to get consistent results more easily is to go for a set of colour matched cini lenses and if you are using multiple cameras make sure they're all the same model of camera or at least cameras which gave the same colour responses.

On mp counts, one advantage of a higher mp count is that you stand a higher chance of being able to crop and still end up with a large image. This has obvious advantages including being able to use smaller lighter and cheaper kit and still being able to make a subject larger in the final frame. You can't always move closer to your subject and even if you can foot zoom alters perspective.

Indeed, and although I didn't find any image quality advantage from the Z8 over the Zf for my 20X16 print tests, the Z8 (along with some technology advantages) is still the obvious choice for my wildlife photography where I am likely to be cropping in.

 

[sk66]

On mp counts, one advantage of a higher mp count is that you stand a higher chance of being able to crop and still end up with a large image.

As a standalone engineering consideration, there are no negatives to increased sensor resolution... but of course, it is not a standalone consideration.

This has obvious advantages including being able to use smaller lighter and cheaper kit and still being able to make a subject larger in the final frame.

This does not correlate IMO. As "smaller/lighter/cheaper" implies smaller sensors, and that has the entirely opposite effect.

You can't always move closer to your subject and even if you can foot zoom alters perspective.

This is true; but if you crop in post, or use a crop sensor/crop mode, you are discarding light. And in terms of image quality, the total light recorded is probably the biggest factor... this is why larger sensors *always win in any comparison.

(*barring large differences in technology)

 

[woof woof]

This does not correlate IMO. As "smaller/lighter/cheaper" implies smaller sensors, and that has the entirely opposite effect.


This is true; but if you crop in post, or use a crop sensor/crop mode, you are discarding light. And in terms of image quality, the total light recorded is probably the biggest factor... this is why larger sensors *always win in any comparison.

Well no, a smaller camera does not necessarily imply a smaller sensor. For example my FF Sony A7cII is smaller than my FF Sony A7III and both are smaller than some APS-C cameras from other brands such as for example some of the Fuji SLR style cameras.

When looking at bulk and weight you really have to look at specific camera and lens combinations but generally on lens size I'm sure you'll agree that one lens can be bigger than another lens of the same focal length and in fact some FF lenses can be smaller than APS-C lenses both within brands and from other brands and that lenses of different focal length vary in size both within and across brands. You don't need imagination or the lenses in your hand for this as there are size comparison tools on line. Assuming you're not looking for razor thin DoF as the max aperture available can affect lens size (but also compare f1.2 lenses of the same focal length and you'll see that some are significantly smaller than others) you could take an A7cII, which is 33mp, and a small shorter focal length lens, take a picture with it and crop it to look like it was taken with larger longer focal length lens.

Telephoto isn't my thing but at the longer end of things lenses do get expensive and heavier and it might be possible to take a picture with a cheaper and lighter shorter focal length lens on a higher mp count camera, crop it and end up with an acceptably large image without using a bigger, heavier and more expensive longer focal length lens. Birds and wildlife aren't my thing but this is an area in which cropping could well be a handy tool, the fact that the kit can be smaller and lighter and cheaper because you're not using that massive, heavy and expensive tele are bonus features.

On image quality, theory is one thing and what is acceptable in practice and in reality is another. What I can tell you is that with some cameras I've had I could crop 100% and produce an acceptable (to me) print, and by acceptable I mean in both image quality and size. Horses for courses but a bigger format sensor with a higher mp count may still produce a higher image quality picture after cropping than APS-C system can. IMO.

 

[sk66]

Well no, a smaller camera does not necessarily imply a smaller sensor. For example my FF Sony A7cII is smaller than my FF Sony A7III and both are smaller than some APS-C cameras from other brands such as for example some of the Fuji SLR style cameras.

If you can get notably smaller/lighter w/o compromising sensor size, then great.

What I can tell you is that with some cameras I've had I could crop 100% and produce an acceptable (to me) print, and by acceptable I mean in both image quality and size.

At some point you get to what I call "the circle of equivalence" where every choice is a tradeoff, and no choice is particularly better than another. E.g. using a longer/slower lens, adding a TC, using a crop sensor/mode, or cropping in post... all can generate similar results with minimal/negligible differences.

 

[woof woof]

If you can get notably smaller/lighter w/o compromising sensor size, then great.


At some point you get to what I call "the circle of equivalence" where every choice is a tradeoff, and no choice is particularly better than another. E.g. using a longer/slower lens, adding a TC, using a crop sensor/mode, or cropping in post... all can generate similar results with minimal/negligible differences.

Clearly you can get small FF cameras the size of or even slightly smaller than SLR style APS-C or even MFT cameras (the Sony RX1 is a rather extreme example) and sometimes the lenses are comparable size wise too. Strange as it seems APS-C and MFT lenses can be the size of FF lenses or near enough as makes no real difference especially if you are willing to take accept smaller aperture on the FF lenses, for eg f2.5/f2.8 on FF rather than f1.8 on MFT etc.

Regarding negligible differences I don't think that's quite the case as a larger format will likely exceed what a smaller format can provide IQ wise assuming the tech generations are near enough the same.

My original point being that cropping a high MP count image taken with a larger format such as FF (bigger ones are available) whilst retaining acceptable IQ may well be possible and if you are willing to do this you may well be able to do it with relatively small bodies (like the Sony range) and shorter focal length lenses thus saving bulk, weight and potentially cost too. This is my opinion after owning 12, 16 and 20mp MFT, 6, 8 and 24mp APS-C and 12, 24 and 33mp FF cameras. Others are free to disagree as what IQ we will accept will vary and ghat affects how much we're willing to crop.

 

[sk66]

Regarding negligible differences I don't think that's quite the case as a larger format will likely exceed what a smaller format can provide IQ wise assuming the tech generations are near enough the same.

That's the thing... larger formats only provide "better" if you use all it. If you degrade it's comparable capability then not necessarily/so
much (i.e. cropping in post)

 

[GAS man]

Fascinating thread, thanks! I have always heard that Leica cameras are the pinnacle of sharpness (I can’t afford one myself! ), but I would be interested in a comparison of the GFX compared to a top Leica camera.

 

[Dave Canon]

You can save a lot of effort by visiting dxomark
I certainly used this data to choose my last camera.

Dave

 

[myotis]

Fascinating thread, thanks! I have always heard that Leica cameras are the pinnacle of sharpness (I can’t afford one myself! ), but I would be interested in a comparison of the GFX compared to a top Leica camera.

In terms of my 20"x16" prints, sharpness between the different cameras wasn't an issue.

This is not the case if you "pixel peep" on the computer, where differences in sharpness are more obvious.

I'm not entirely sure that Leica lenses are seen as the pinnacle of sharpness; rather, their reputation is more about the way they render colour and tonality (of which resolution is a factor).

As it so happens, as I would love a Leica Q3 (to allow me to switch between zone-focussing and auto-focusing,) I downloaded a few sample raw files from a Leica Q3 review and I didn't think there was anything special about the sharpness (pixel peeping on the computer screen). But of course, this isn't one of Leica's best lenses, and it wasn't a proper comparison.

But I would still put my money on the GFX being better, especially if you were using a GFX 100 rather than my GFX50.

Not all will agree but, as a generalisation, and all things being equal, I would expect a bigger sensor as coming with the potential to always outperform a smaller sensor, even if lens quality still plays a part. But the GFX lenses have a good reputation.

My tests here had the sharpness required capped at the sharpness needed to make 20 x 16 prints, so it's not possible to make any assessments about "pinnacles" of performance between the different sensor sizes/lenses used.

 

[myotis]

You can save a lot of effort by visiting dxomark
I certainly used this data to choose my last camera.

Dave

The other useful site is ( but this assumes one or both list the cameras and lenses you are interested in)

Photons to Photos

But these sites don't save me any effort.

The main reason for going through the tedium of tests like the one described here is to learn about the capabilities and limitations of the equipment that I already own, in the context of how I use that equipment.

I see this as an important part of learning my craft as a photographer.

When I worked in professional photography (in the film days), whenever we had downtime, it was filled with making tests of some sort.

 

[Dave Canon]

That's the thing... larger formats only provide "better" if you use all it. If you degrade it's comparable capability then not necessarily/so
much (i.e. cropping in post)

A good point which is why I choose my cameras by their potential capability. Though I do not set out to have large crops, sometimes it is unavoidable.

Dave