Interesting comparison of FF and medium format

[DanK]

I found this interesting, and guessed wrong as much as right. It would be better to see large prints, of course, but still:

https://www.lensrentals.com/blog/202...-a-modern-one/

[xpatUSA]

I found this interesting, and guessed wrong as much as right. It would be better to see large prints, of course, but still:

https://www.lensrentals.com/blog/202...-a-modern-one/

Interesting link, thanks.

Not a rebuttal but, while the side-by-side images showed significant differences, if I only owned the one or the other camera and if I only viewed the output fit-to-screen on my 24" NEC - I doubt that I would be bothered much, if at all.

[DanK]

Interesting link, thanks.

Not a rebuttal but, while the side-by-side images showed significant differences, if I only owned the one or the other camera and if I only viewed their output fit-to-screen on my 24" NEC - I doubt that I would be bothered much, if at all.

That's part of their point: displaying on screen, it doesn't make much differerence. It would have been interesting for them to have linked to full-size images. But I think their larger point is diminishing returns.

[xpatUSA]

That's part of their point: displaying on screen, it doesn't make much differerence. It would have been interesting for them to have linked to full-size images. But I think their larger point is diminishing returns.

Agreed and understood.

[Manfred M]

I was lucky enough to know a couple of people that own the FujiFilm GFX100s and one with the Hasselblad 907X, so have had access to the cameras and the images that they produce.

From an image quality standpoint the Fuji appears to create stronger images (the Hasselblad is 50MP versus the FuujiFilm being 100 MP). The colours are much more subtle and it is noticeable, even in the images shown in the article that Dan posted. It is even more obvious when looking at medium sized prints.

[DanK]

From an image quality standpoint the Fuji appears to create stronger images (the Hasselblad is 50MP versus the FuujiFilm being 100 MP). The colours are much more subtle

So why might that be?

--There is no difference in sensor size between the two.

--The Fuji has smaller photosites, not something one generally associates with higher image quality.

--The Fuji has more pixels, but I would expect that to affect fine detail, not color.

The latter two affect the comparison between the 5D III and the Fuji as well. The pixel pitch of the 5D III is 6.2 microns, while that of the Fuji is 3.8.

Wouldn't the subtlety of colors likely reflect something else, for example, the type of sensor (Beyer vs. x-Trans), other aspects of the sensor or internal circuitry, or postprocessing? If that's the case, it's not really relevant to the MF/FF question.

[tclune]

I also got them all, though I must admit that it was only the shots with a person in them that I found obvious.

[xpatUSA]

From an image quality standpoint the Fuji appears to create stronger images (the Hasselblad is 50MP versus the FuujiFilm being 100 MP). The colours are much more subtle

So why might that be?

Wouldn't the subtlety of colors likely reflect something else, for example, the type of sensor ([Bayer] vs. x-Trans), other aspects of the sensor or internal circuitry, or post-processing? If that's the case, it's not really relevant to the MF/FF question.

Gents: "Subtle" is not one of my favorite words in the lexicon of photography ...

... How is the subtlety of a color measured, I wonder.

[Manfred M]

So why might that be?

--There is no difference in sensor size between the two.

--The Fuji has smaller photosites, not something one generally associates with higher image quality.

--The Fuji has more pixels, but I would expect that to affect fine detail, not color.

The latter two affect the comparison between the 5D III and the Fuji as well. The pixel pitch of the 5D III is 6.2 microns, while that of the Fuji is 3.8.

Wouldn't the subtlety of colors likely reflect something else, for example, the type of sensor (Beyer vs. x-Trans), other aspects of the sensor or internal circuitry, or postprocessing? If that's the case, it's not really relevant to the MF/FF question.

From what the "experts" tell me is that they suspect it is that the sensor is a true 16-bit design rather than a 14-bit that has been packed to 16-bits. Neither are xTrans sensor; both are Sony Bayer sensors. All I can say is I saw two side-by-side 17 x 22 prints and the difference was definitely visible. Same subject, same lens, both shot from a tripod, same processing, etc. The camera body was the only difference.

[xpatUSA]

From what the "experts" tell me is that they suspect it is that the sensor is a true 16-bit design rather than a 14-bit that has been packed to 16-bits.

I wonder what the "experts" meant by "a 14-bit that has been packed to 16-bits"?

Two of several ways to go:

1) The 14-bit number gets a couple of most significant bits added with a value of zero which does not alter the value of the number.

2) The 14-bit number gets re-scaled by a factor of 65536/16384 (decimal) which does alter the value of the number.

The difference is not subtle but does offer plenty of room for obfuscation ...

[DanK]

From what the "experts" tell me is that they suspect it is that the sensor is a true 16-bit design rather than a 14-bit that has been packed to 16-bits.

Interesting. I don't understand it either, but it isn't an attribute of MF per se, so it's irrelevant to the general MF/FF question.

Note also that they used a 22.3 MP FF sensor in a camera introduced almost exactly 10 years ago. Another major way in which this is not a minimal contrast. It would have been a much cleaner comparison if they had used a X9 or R3.

But then again, their point wasn't really MF vs. FF. Their argument was that even comparing a current MF to a FF that is literally 4 generations out of date in the Canon line-up, the differences are small.

[xpatUSA]

From what the "experts" tell me is that they suspect it is that the sensor is a true 16-bit design rather than a 14-bit that has been packed to 16-bits.

Interesting. I don't understand it either, but it isn't an attribute of MF per se, so it's irrelevant to the general MF/FF question.

Agreed. And, since 14-bit and 16-bit imply ADCs, the ADC may not even be part of the "sensor design", i.e. was not located thereon!

[Manfred M]

I wonder what the "experts" meant by a 14-bit that has been packed to 16-bits?

Two of several ways to go:

1) The 14-bit number gets a couple of most significant bits added with a value of zero which does not alter the value of the number.

2) The 14-bit number gets re-scaled by a factor of 65536/16384 (decimal) which does alter the value of the number.

The difference is not subtle but does offer plenty of room for obfuscation ...

Ted -my understanding is that the output (after A-D conversion) is a 2^14 value. Packing it makes turns the number into a 2^16 representation without actually changing the numeric value. A true 16 bit sensor outputs 2^16 data, so there is more data to use.

[xpatUSA]

Originally Posted by xpatUSA I wonder what the "experts" meant by a 14-bit that has been packed to 16-bits?

Two of several ways to go:

1) The 14-bit number gets a couple of most significant bits added with a value of zero which does not alter the value of the number.

2) The 14-bit number gets re-scaled by a factor of 65536/16384 (decimal) which does alter the value of the number.

Ted -my understanding is that the output (after A-D conversion) is a 2^14 value.

Which makes the ADC a 14-bit model ...

Packing it makes turns the number into a 2^16 representation without actually changing the numeric value.

... so, the elusive "packing" means adding two MSBs to the 14-bit ADC output per my 1) above.

A true 16 bit sensor outputs 2^16 data, so there is more data to use.

Which tells us that the ADC is on the sensor in that case.

Thanks for your understanding, Manfred.

P.S. a bit of searching reveals that the term "packing" is used more in data-compression, as in "bit packing" rather than changing a binary number format: so perhaps the term is less appropriate in this discussion ...

P.P.S. also found it in reference to the Binary Coded Decimal format but still not really applicable to fitting 14-bit data into a 16-bit space.

[Manfred M]

Interesting. I don't understand it either, but it isn't an attribute of MF per se, so it's irrelevant to the general MF/FF question.

Note also that they used a 22.3 MP FF sensor in a camera introduced almost exactly 10 years ago. Another major way in which this is not a minimal contrast. It would have been a much cleaner comparison if they had used a X9 or R3.

But then again, their point wasn't really MF vs. FF. Their argument was that even comparing a current MF to a FF that is literally 4 generations out of date in the Canon line-up, the differences are small.

I agree that the weakness of the article was to introduce a different format into the mix, but there is no "clean" comparisons across different generations of sensors. Showing the GFX 100s, which is considered to be one of the highest IQ cameras on the market today (excluding the really high end Hasselblad and Phase One offerings) shows how well the 10 year old Canon holds up today.

What surprised me more that a little is how easily the higher IQ was noticeable in the down-sampled, sRGB files that were being shown. I have generally looked at medium sized prints as a reference, due to the wider colour space that is generally used when printing.

[pnodrog]

It's possible that an expert used the term 14bits packed to 16bits. It would be far more correct to say 14bits packed into 16bits. Packing is a term used when you shift (or rearrange) the data across normal boundaries (8bit, 16bit word, long word etc) to make use of any unallocated bits. It saves memory space but adds processing time both to pack and unpack.

The real issue would be whether it was a 14 or 16 bit A/D conversion. The format of storage is almost irrelevant.

[pnodrog]

Here is an illustration of what is meant by packing. The data types can vary as can the packing methods. The illustration uses the fairly common shift to pack method.



Probably just another bit of useless information that a photographer should not worry about.

[xpatUSA]

Here is an illustration of what is meant by packing. The data types can vary as can the packing methods. The illustration uses the fairly common shift to pack method.



Probably just another bit of useless information that a photographer should not worry about.

Thanks for the clear illustration of "packing"!

From which one could deduce that 14-bit data packed into words does not gain much space but would, as you said, require a processing burden in a genre that is more concerned with speed of processing than file size.

Personally, I could envisage a 14-bit ADC connected to a 16-bit data bus with the 2 MSB's grounded somewhere.

Agreed that none of us should care, but - you know me!

[DanK]

Back to the more fundamental question: what advantage is there in going from 14 to 16 bits?

The human eye, if I remember correctly, can differentiate something like 12 or 15 million colors.

Here are the number of colors corresponding to various bit depths:

8: 16.8 million
10: 1.07 billion. If I'm not mistaken, this is the bit depth of the LUT in high-end wide-gamut monitors
12: 68 billion
14: 4 trillion
16: 35 trillion

I understand (I think) that one needs more than 8 bits to encompass the gamut visible to people, but is there reason to expect that increasing from 4 to 35 trillion colors will produce differences that are apparent to the naked eye?

There must be some empirical data about this.

[xpatUSA]

Back to the more fundamental question: what advantage is there in going from 14 to 16 bits?

The human eye, if I remember correctly, can differentiate something like 12 or 15 million colors.

Here are the number of colors corresponding to various bit depths:

8: 16.8 million
10: 1.07 billion. If I'm not mistaken, this is the bit depth of the LUT in high-end wide-gamut monitors
12: 68 billion
14: 4 trillion
16: 35 trillion

I understand (I think) that one needs more than 8 bits to encompass the gamut visible to people, but is there reason to expect that increasing from 4 to 35 trillion colors will produce differences that are apparent to the naked eye?

There must be some empirical data about this.

As to exact numbers of bits, a slight complication is that ADCs have a parameter called "effective number of bits" (ENOB).

https://en.wikipedia.org/wiki/Effective_number_of_bits

For example, the 12-bit ADC model used in my Sigma SD9 has a 10.5-bit ENOB according to Analog Devices Inc.

Even worse, raw data can get messed with in-camera. For example, a saturated sensor pixel does not mean that what is written to the card for that pixel is exactly the ADC max output. I had one Sigma where the saturation showed as about 6,000 for a 12-bit ADC (4095 max). Another Sigma model showed about 3,500 for the same sensor and ADC. But, in terms of empirical data, I have found that no differences were apparent to the naked eye even at such low "bit depths", all other things being equal. Therefore, I think that increasing from 4 to 35 trillion colors will make no visible difference.