Luminous Landscape Forum
Equipment & Techniques => Medium Format / Film / Digital Backs – and Large Sensor Photography => Topic started by: wallpaperviking on January 15, 2017, 02:59:09 am
-
Why does the Hasselblad X1D appear listed as having 16-Bit color in various spec sheets whereas the Fuji GFX is still only "rumoured" to have 16 Bit (quite possibly 14 bit) color?
I would have thought sharing the exact same sensor would have put them in the same league together?
-
I would have thought sharing the exact same sensor would have put them in the same league together?
Hi,
There is more to a camera than the sensor alone. Supporting electronics (such as the Analog to Digital Converter (ADC)) and the file format and Raw conversion need to be adapted too.
If done well, then a true 16-bit pipeline will give superior results (like the Phase One IQ3 100 does).
If and how well that's implemented in the Hasselblad and Fuji offerings remains to be seen. When Raw files from production cameras become available then an objective analysis will become possible. RawDigger is usually pretty fast with delving into the specifics of the file peculiarities, also because the Raw conversion libraries are used for Fast Raw Viewer, which should be useful with such large file sizes.
Cheers,
Bart
-
Perhaps Fuji just don't bother with it. 16-bit is more of a marketing scheme. Technically these Sony CMOS sensors, while being the state-of-the-art in terms of dynamic range, only has no more than 14 stops of dynamic range at pixel level, and there is no practical advantage to have more than 14-bit encoding in their raw files. The CCD sensors of course cannot benefit from 16-bit raw files given that they even have less dynamic range than these Sony CMOS sensors.
-
Hi,
The X1D is a 14 bit device. Hasselblad actually states this quite clearly. 16-bit colour is marketing nonsense.
The 100 MP sensor used in the Phase One IQ3-100 is in fact a 16 bit device delivering 15 bits worth of data.
I enclose a raw digger histogram, it has peaks for each fourth channel. The in between channels having data indicates that some processing was done on the data before writing the raw file.
Best regards
Erik
Why does the Hasselblad X1D appear listed as having 16-Bit color in various spec sheets whereas the Fuji GFX is still only "rumoured" to have 16 Bit (quite possibly 14 bit) color?
I would have thought sharing the exact same sensor would have put them in the same league together?
-
Hi,
The X1D is a 14 bit device. Hasselblad actually states this quite clearly. 16-bit colour is marketing nonsense.
Hi Erik,
I'm a bit more cautious, given my earlier experience with the early samples of the Phase One IQ3 100. The early samples were shot in 14-bit mode, but later examples became available in 16-bit mode. I do not know if the Hasselblad will offer such a setting (and the firmware seems to be a works in progress).
The 100 MP sensor used in the Phase One IQ3-100 is in fact a 16 bit device delivering 15 bits worth of data.
As we found out later, there is a special setting that delivers a different raw format to accommodate that 16-bit data pipeline. My measurements (http://forum.luminous-landscape.com/index.php?topic=106969.msg882537#msg882537) indicated an increase from 13.65 stops at nominal ISO 50 in 14-bit mode to a DR of 14.39 stops in 16-bit mode. Not quite the claimed 15-bits but 14.36 stops of engineering DR is still very high (given that one usually loses at least 0.5 stop to residual noise sources), and a significant enough improvement albeit at the cost of (even) larger filesizes.
I enclose a raw digger histogram, it has peaks for each fourth channel. The in between channels having data indicates that some processing was done on the data before writing the raw file.
Yes, although I'll reserve my final analysis for regular production models with up-to-date firmware.
These are interesting times ...
It will also be interesting to see how the available lens ranges evolve, and how well available lenses with adapters will match a Digital sensor at the given flange distances and given filter stacks (if any). Lenses designed for digital sensors are different from analog/film designs, and dedicated lenses for mirrorless cameras are most likely different as well.
Cheers,
Bart
-
Hi Bart,
I have been involved in the IQ3100 SNAFU…
But, all implementations of the 44x33 mm sensor have been 14 bits. Hasselblad claims up to 14 EV DR, which is a clear indication that there is not more than 14 bits worth of data.
Once you get beyond 14 EV of DR, you would need 15 bits of data to represent that dynamic range.
Many photographers quote "16 bit colour" as an advantage of MFD, but Photoshop even cannot handle 16 bit of colour unless it is used in 32-bits or floating point mode.
As I can see it, 16 bit colour is just marketing speak. Phase one even used to employ a 14 bit file format. That was the cause of the IQ3 100MP SNAFU. Old versions of RawDigger interpreted the data in the old 14-bit format.
Anders Torger has discovered that the IQ3 100 MP files actually were using 16 bits instead of the original 14 bits.
Now, using 14 bit data paths on the older backs, including the IQ 350, was sound engineering judgement, but calling those data 16 bit files was simply a lie. A marketing lie, but still a lie.
The IQ3 100 MP delivers beyond 14 bits, so Phase One modified their pipeline to 16 bits, good.
I am surprised to see that the IQ3 100 MP delivers past 14 bits, but it seems to be the case and I have no objection with it.
Best regards
Erik
Hi Erik,
I'm a bit more cautious, given my earlier experience with the early samples of the Phase One IQ3 100. The early samples were shot in 14-bit mode, but later examples became available in 16-bit mode. I do not know if the Hasselblad will offer such a setting (and the firmware seems to be a works in progress).
As we found out later, there is a special setting that delivers a different raw format to accommodate that 16-bit data pipeline. My measurements (http://forum.luminous-landscape.com/index.php?topic=106969.msg882537#msg882537) indicated an increase from 13.65 stops at nominal ISO 50 in 14-bit mode to a DR of 14.39 stops in 16-bit mode. Not quite the claimed 15-bits but 14.36 stops of engineering DR is still very high (given that one usually loses at least 0.5 stop to residual noise sources), and a significant enough improvement albeit at the cost of (even) larger filesizes.
Yes, although I'll reserve my final analysis for regular production models with up-to-date firmware.
These are interesting times ...
It will also be interesting to see how the available lens ranges evolve, and how well available lenses with adapters will match a Digital sensor at the given flange distances and given filter stacks (if any). Lenses designed for digital sensors are different from analog/film designs, and dedicated lenses for mirrorless cameras are most likely different as well.
Cheers,
Bart
-
Erik can you clarify/explain the 32-bit or floating point mode point? Thanks!
-
The X1D is a 14 bit device. Hasselblad actually states this quite clearly. 16-bit colour is marketing nonsense.
Where does Hasselblad say this?
I read on the website:
Colour definition 16 bit; Dynamic range up to 14 stops
How does colour depth or bits per pixels relate to dynamic range anyway?
All I know is that my 2007 H3DII has 16 bits of this "marketing nonsense" and the colour is far better than my new 14 bit Canon.
Perhaps Fuji has 14 bits because it is a lesser camera, and there is nothing wrong with that for the market they choose.
-
Erik can you clarify/explain the 32-bit or floating point mode point? Thanks!
Also known as a High Dynamic Range image, can contain many more stops of dynamic range.
-
No, I get that. It's the "Photoshop even cannot handle 16-bit color," part that confused me.
-
Hi,
Photoshops internal colour handling is 15 bits.
https://books.google.se/books?id=JrNmCgAAQBAJ&pg=PA330&lpg=PA330&dq=photoshop+colour+handling+15+bits&source=bl&ots=nga29uBAlF&sig=c7cWj-PWYoazDpXbFBUZvyJ8fjk&hl=sv&sa=X&ved=0ahUKEwiM_8GZg8bRAhUC_SwKHekiDd8Q6AEISDAF#v=onepage&q=photoshop%20colour%20handling%2015%20bits&f=false
Best regards
Erik
No, I get that. It's the "Photoshop even cannot handle 16-bit color," part that confused me.
-
Hi,
This shows that desinformation works.
Dynamic range is range between highest reproducible white and lowest reproducible dark. Darks are always limited by noise.
Light, on it's own, is noisy. It is coming in quanta. A pixel can capture so and so many photons (quanta of light) your sensor on the Hassy can hold around 60000 electron charges per pixel. Normally, an electron charge corresponds to a captured photon.
As I said, the light itself is noisy. If you would photograph a white piece of paper overexposing 3 EV for the white you would have around ETTR (Expose To The Right) exposure with the sensor fully utilised. In this the paper white would be represented by a bell curve, centered around 60000 with 65 of the pixels having between 59750 and 60250 electron charges.
Now, looking at the other end the Hasselblad sensor has a readout noise of around 10.5 e-, if I recall correctly. So the ratio between the Full Well and the readout noise would be 60000 / 10.5 -> 6190. If we will represent that range with a binary number it would need log(6190) / log (2) -> 12.5 bits. So that sensor is a barely 13 bit device.
To explain the difference between your Canon and Hasselblad you need to go somewhere else. One factor is that your Canon has fewer pixels and Canon has quite noisy readout, too. If you look at actual pixels you would see the noise at the pixel level. But at any other size the Hasselblad would use more pixels for each visible dot. That would reduce the noise. But not because it has more bits of data but because it has more pixels.
The major factors for colour rendition are in all probability two. One is the Color Filter Array. The pixels don't have colour, they just count photons. Colour is added by using a Color Filter Array.
The Kodak CCDs used to have quiet extreme CFA designs, probably optimised for studio light. Canon's are often used in mixed light and under poor light conditions. Their CFA designs are more general purpose and probably more permissive.
The other factor is the colour profile. The signal coming out of the sensor is a voltage, that is amplified and converted to binary data by an ADC, which happens to be a commercial flash converter older cameras.
The binary data is converted to RGB data using a colour profile. Hasselblad is known for well made colour profiles.
Later generation CMOS sensors have moved the ADCs to the sensor. There is normally thousands of them, one for each column. This allows to use slow but precise converters. But it also means that the vendor of the sensor decides the number of bits and the Sony sensor used by Fuji and Hasselblad is a 14-bit device.
The enclosed screen dump of the RawDigger histogram of a raw file from the X1D-50c clearly indicates that each fourth channel contains data. That the gaps also contain data indicates that some image manipulation has been done on the raw data. That could be flat field correction of the raw data, or simply some dither noise added.
The second screen dump is coming from an Phase One IQ350 shot, courtesy of Digital Transitions. This shows no sign of postprocessing in raw. Phase One also claims 16 bit colour. The IQ 350 uses the same Sony sensor as the X1D-50c
The last attachments shows the white patch on IQ 350 shot. It shows what a great spread of signal a single patch can have. Some of that may come from the rough surface of the patch, but most is coming from variation of incident photons.
Best regards
Erik
Where does Hasselblad say this?
I read on the website:
Colour definition 16 bit; Dynamic range up to 14 stops
How does colour depth or bits per pixels relate to dynamic range anyway?
All I know is that my 2007 H3DII has 16 bits of this "marketing nonsense" and the colour is far better than my new 14 bit Canon.
Perhaps Fuji has 14 bits because it is a lesser camera, and there is nothing wrong with that for the market they choose.
-
Where does Hasselblad say this?
I read on the website:
Colour definition 16 bit; Dynamic range up to 14 stops
How does colour depth or bits per pixels relate to dynamic range anyway?
All I know is that my 2007 H3DII has 16 bits of this "marketing nonsense" and the colour is far better than my new 14 bit Canon.
Perhaps Fuji has 14 bits because it is a lesser camera, and there is nothing wrong with that for the market they choose.
When you bracket two images by 1EV and open these two images with RawDigger, you would be able to notice that the levels around the same region in the frame are about 2:1 ratio between these two images. This means every 1 extra bit stores one extra stop of dynamic range. If the sensor can only capture 14 stops dynamic range, then 16-bit raw file would be overkill, and the lowest 2 bits would be just noise (useless information stored).
Your H3DII is a CCD sensor and has less than 12 stops of dynamic range. That means a raw file with more than 12-bit would be overkill for the H3DII, and anything beyond 12-bit is just for marketing/metaphysics without technical benefit.
-
Hi,
This shows that desinformation works.
Dynamic range is range between highest reproducible white and lowest reproducible dark. Darks are always limited by noise.
Light, on it's own, is noisy. It is coming in quanta. A pixel can capture so and so many photons (quanta of light) your sensor on the Hassy can hold around 60000 electron charges per pixel. Normally, an electron charge corresponds to a captured photon.
As I said, the light itself is noisy. If you would photograph a white piece of paper overexposing 3 EV for the white you would have around ETTR (Expose To The Right) exposure with the sensor fully utilised. In this the paper white would be represented by a bell curve, centered around 60000 with 65 of the pixels having between 59750 and 60250 electron charges.
Now, looking at the other end the Hasselblad sensor has a readout noise of around 10.5 e-, if I recall correctly. So the ratio between the Full Well and the readout noise would be 60000 / 10.5 -> 6190. If we will represent that range with a binary number it would need log(6190) / log (2) -> 12.5 bits. So that sensor is a barely 13 bit device.
To explain the difference between your Canon and Hasselblad you need to go somewhere else. One factor is that your Canon has fewer pixels and Canon has quite noisy readout, too. If you look at actual pixels you would see the noise at the pixel level. But at any other size the Hasselblad would use more pixels for each visible dot. That would reduce the noise. But not because it has more bits of data but because it has more pixels.
The major factors for colour rendition are in all probability two. One is the Color Filter Array. The pixels don't have colour, they just count photons. Colour is added by using a Color Filter Array.
The Kodak CCDs used to have quiet extreme CFA designs, probably optimised for studio light. Canon's are often used in mixed light and under poor light conditions. Their CFA designs are more general purpose and probably more permissive.
The other factor is the colour profile. The signal coming out of the sensor is a voltage, that is amplified and converted to binary data by an ADC, which happens to be a commercial flash converter older cameras.
The binary data is converted to RGB data using a colour profile. Hasselblad is known for well made colour profiles.
Later generation CMOS sensors have moved the ADCs to the sensor. There is normally thousands of them, one for each column. This allows to use slow but precise converters. But it also means that the vendor of the sensor decides the number of bits and the Sony sensor used by Fuji and Hasselblad is a 14-bit device.
The enclosed screen dump of the RawDigger histogram of a raw file from the X1D-50c clearly indicates that each fourth channel contains data. That the gaps also contain data indicates that some image manipulation has been done on the raw data. That could be flat field correction of the raw data, or simply some dither noise added.
The second screen dump is coming from an Phase One IQ350 shot, courtesy of Digital Transitions. This shows no sign of postprocessing in raw. Phase One also claims 16 bit colour. The IQ 350 uses the same Sony sensor as the X1D-50c
The last attachments shows the white patch on IQ 350 shot. It shows what a great spread of signal a single patch can have. Some of that may come from the rough surface of the patch, but most is coming from variation of incident photons.
Best regards
Erik
Exceptional explanation. Thank you for taking the time to write it!
-
Erik - Another great example of why, "When Erik speaks, I listen and don't question"!
Well explained my Ole Friend!
Jack
-
There is more to a camera than the sensor alone.....
Cheers,
Bart
+1
-
The sensor used by both these cameras does ADC onboard, outputting a digital signal, so I very much doubt that there is a difference in the number of _significant_ bits in their raw output. If indeed Hasselblad delivers 16 bits while Fujiifilm delivers only 14, it would be due to Hasselblad padding 14 bit sensor output with two zeroes to fit its previous practice of delivering 16 bit numbers (two full bytes), without their being more than 14 significant bits.
-
The sensor used by both these cameras does ADC onboard, outputting a digital signal, so I very much doubt that there is a difference in the number of _significant_ bits in their raw output. If indeed Hasselblad delivers 16 bits while Fujiifilm delivers only 14, it would be due to Hasselblad padding 14 bit sensor output with two zeroes to fit its previous practice of delivering 16 bit numbers (two full bytes), without their being more than 14 significant bits.
Now that's a disturbing thought!
-
Why? It has bees done by Hasselblad and phase one before.
Gesendet von iPhone mit Tapatalk
-
Why? It has bees done by Hasselblad and phase one before.
Gesendet von iPhone mit Tapatalk
Peronally, I'm not sure whether it has been done by Phase or HB before is the point; I think the point that BJL was making is whether it actually has any practical significance.
-
Perhaps wrong wording. I don't know what they are doing or not, but I think it was shown by much smarter people than me, that the only sensor actually needing more than 14bits is the 100MP version.
Gesendet von iPhone mit Tapatalk
-
This title better describes what might be happening with 2 cameras using the same "base" Sony sensor. If later someone has BOTH cameras then we will know which company is true to their word.
-
Perhaps wrong wording. I don't know what they are doing or not, but I think it was shown by much smarter people than me, that the only sensor actually needing more than 14bits is the 100MP version.
Gesendet von iPhone mit Tapatalk
That could be true, or it could be a marketing play...
-
Now that's a disturbing thought!
Photography files are big and data paths slow, so camera makers have avoided using unnecessary bits in their files. In audio there is no such thing as 24 bit (real) signal with claimed 144 dB (or actually close to 147 dB) of dynamic range, as the very best converters costing thousands (even per channel) can manage about 127 dB of dynamic range which corresponds to 21 bits only. Typical cheap recorders are happy to reach 16 bits of real accuracy when set to 24 bits. Still 24 bit quality is claimed by all, cheapest being the loudest. This is because 16 and 24 bits are standards and everybody has to adhere to those and fill the missing bits with marketing talk. Still some uninformed (professionals even) ask for 32 bit "resolution", which would require the systems to run all the way through on over 10 kV rail levels to avoid thermal noise at the other end...
Back to photography...
-
Considering this tread was last posted to on Jan. 31, 2017, here is an update for a file shot on the X1D.
Production units are 16 bit:
Make
Hasselblad
Camera Model Name
Hasselblad X1D-50c
Orientation
Rotate 270 CW
X Resolution
72
Y Resolution
72
Planar Configuration
Chunky
Resolution Unit
None
Software
v1.15.0-6541
Modify Date
2017:04:20 12:49:23
Image Width
8384
Image Height
6304
Bits Per Sample
16
-
Hi Jack,
As far as we know, the 44x33mm sensor is a 14 bit device. I have checked out the raw image used by the DPReview in their image comparator and it is certainly a 16 bit file, but it has all signs of only having 14 bits data, as only each fourth channel seems to contain real data. But it also seems that some calculations/scaling has been done on the data.
So, I would say that it is a sixteen bit file containing fourteen bits worth of data.
Best regards
Erik
Considering this tread was last posted to on Jan. 31, 2017, here is an update for a file shot on the X1D.
Production units are 16 bit:
Make
Hasselblad
Camera Model Name
Hasselblad X1D-50c
Orientation
Rotate 270 CW
X Resolution
72
Y Resolution
72
Planar Configuration
Chunky
Resolution Unit
None
Software
v1.15.0-6541
Modify Date
2017:04:20 12:49:23
Image Width
8384
Image Height
6304
Bits Per Sample
16
-
Erik,
The Sony I have is 14 bits, and the files from the Hassie H5D 50c WiFi are FAR superior to the Sony in color ramp.
Sony is sharp, good resolution but it is so easy to see the difference in the smoothness of gradations - sky, etc.
Jack
PS: Hope you are doing well my Friend! All is well here.
-
Hi Erik,
Maybe you can work fixed pattern noise into your explanation. The data from the sensor is never really "raw.
Edmund
Hi,
This shows that desinformation works.
Dynamic range is range between highest reproducible white and lowest reproducible dark. Darks are always limited by noise.
Light, on it's own, is noisy. It is coming in quanta. A pixel can capture so and so many photons (quanta of light) your sensor on the Hassy can hold around 60000 electron charges per pixel. Normally, an electron charge corresponds to a captured photon.
As I said, the light itself is noisy. If you would photograph a white piece of paper overexposing 3 EV for the white you would have around ETTR (Expose To The Right) exposure with the sensor fully utilised. In this the paper white would be represented by a bell curve, centered around 60000 with 65 of the pixels having between 59750 and 60250 electron charges.
Now, looking at the other end the Hasselblad sensor has a readout noise of around 10.5 e-, if I recall correctly. So the ratio between the Full Well and the readout noise would be 60000 / 10.5 -> 6190. If we will represent that range with a binary number it would need log(6190) / log (2) -> 12.5 bits. So that sensor is a barely 13 bit device.
To explain the difference between your Canon and Hasselblad you need to go somewhere else. One factor is that your Canon has fewer pixels and Canon has quite noisy readout, too. If you look at actual pixels you would see the noise at the pixel level. But at any other size the Hasselblad would use more pixels for each visible dot. That would reduce the noise. But not because it has more bits of data but because it has more pixels.
The major factors for colour rendition are in all probability two. One is the Color Filter Array. The pixels don't have colour, they just count photons. Colour is added by using a Color Filter Array.
The Kodak CCDs used to have quiet extreme CFA designs, probably optimised for studio light. Canon's are often used in mixed light and under poor light conditions. Their CFA designs are more general purpose and probably more permissive.
The other factor is the colour profile. The signal coming out of the sensor is a voltage, that is amplified and converted to binary data by an ADC, which happens to be a commercial flash converter older cameras.
The binary data is converted to RGB data using a colour profile. Hasselblad is known for well made colour profiles.
Later generation CMOS sensors have moved the ADCs to the sensor. There is normally thousands of them, one for each column. This allows to use slow but precise converters. But it also means that the vendor of the sensor decides the number of bits and the Sony sensor used by Fuji and Hasselblad is a 14-bit device.
The enclosed screen dump of the RawDigger histogram of a raw file from the X1D-50c clearly indicates that each fourth channel contains data. That the gaps also contain data indicates that some image manipulation has been done on the raw data. That could be flat field correction of the raw data, or simply some dither noise added.
The second screen dump is coming from an Phase One IQ350 shot, courtesy of Digital Transitions. This shows no sign of postprocessing in raw. Phase One also claims 16 bit colour. The IQ 350 uses the same Sony sensor as the X1D-50c
The last attachments shows the white patch on IQ 350 shot. It shows what a great spread of signal a single patch can have. Some of that may come from the rough surface of the patch, but most is coming from variation of incident photons.
Best regards
Erik
-
I see here a lot of hypothesis and not a lot of confirmed fact. The pictures of the files apparently shows there is data bit in each bit so by definition it is a 16 bit file. Did all 16 bits come from the sensor? Maybe not. The sensor does not produce the raw file, the camera does. Given that the sensor actually does not record colour, but only the brightness of a single colour and then compares that to the brightness of the coloured pixels next to it, then possibly the camera is doing it's own interpretation. The important thing is, who cares.
I ditto the comment that the 14 bit Sony does not have the colour of the 16 bit Hasselblad. My new 14 bit Canon which cost me three times the price I paid for the old Hasselblad also does not have as good a colour. That's all the proof I need. If I can drive to the location and it's not moving I use the Hasselblad.
-
Hi Jack,
Nice to hear that things are well!
Thanks for your observations. But, the difference you see may be caused by some other factors.
- One difference is the size of the sensor, a larger sensor can collect more light. That would give less noise.
- Another difference may be if you use Phocus with the Hasselblad and say Lightroom with the Sony. Just to say, Anders Torger thinks that Hasselblad has licensed parts of the Phocus processing pipeline to Adobe, so that Lightroom uses in part Hasselblad's algorithms for colour processing for Hasselblad images. (*)
- It may be that the CFA (color filter array)on X1D is more balanced for daylight while the A7rII is more general purpose, with higher sensivity for reds and lower sensivity in the blues.
I would also argue that the difference between 16 and 14 bit coding would not matter in highlights. Let me explain…
We can assume that full well capacity on a modern sensor is around 6000 e-, a simplification, but we can assume that corresponds to 6000 photons. Let's say 64000 as it happen to be a convenient number.
Let us also assume that sky is exposed 3EV below saturation. That would be a pretty dark sky :-) So we would have on the average 64000/3^2 -> 8000 photons captured by each pixel. Photon arrival is regarded to be Poisson distributed and standard of deviation for Poisson distributed data is the square root of the number of samples.
Sqrt(8000) is 89.4, so 65% of the samples would be within 7911 - 8089, but 35% would vary even more. So, the data varies something like 180 data numbers. Would we code it with 14 bit, the average would be 2000 with a spread between 1975 and 2025. So noise in captured light dominates many times over precision of data.
Best regards
Erik
(*) Anders Torger has looked into how Phocus handles colours, which is quiet different from Adobe's handling, and he thinks he can see traces from that in Lightroom conversions. I don't have the time to find the reference, and it may be that I have just dreamed it, but I am pretty sure Anders has mentioned it somewhere.
Erik,
The Sony I have is 14 bits, and the files from the Hassie H5D 50c WiFi are FAR superior to the Sony in color ramp.
Sony is sharp, good resolution but it is so easy to see the difference in the smoothness of gradations - sky, etc.
Jack
PS: Hope you are doing well my Friend! All is well here.
-
The sensor is only 14bit. The rest of the 16bit raw format is filled up. It's the same with all CCD and CMOS backs until the IQ100 and Hassi 100 came along. They actually need 16bit or let's say 15bit.
The Sony has bad colours so has the Nikon and canon stuff. However, if you really spent time calibration them you can get way better results.
We can get into the discussion whether interpolating the 14bit data into a 16bit raw has any benefits. From my tests with the GFX it is certainly NOT the case. The IQ350 isn't any different and it had amazing files.
Gesendet von iPhone mit Tapatalk
-
Christopher,
"From my tests with the GFX it is certainly NOT the case. The IQ350 isn't any different and it had amazing files. "
I have been away from the forum for a while for personal reasons and just getting back in. IF you explained in detail what you mean by the above statement else where, please give me a link so I can understand.
If not elsewhere, would you clarify for me here?
Thanks,
Jack
-
We can get into the discussion whether interpolating the 14bit data into a 16bit raw has any benefits. From my tests with the GFX it is certainly NOT the case. The IQ350 isn't any different and it had amazing files.
Assuming the support electronics for the ADC are 16-bit as well, the benefit of quantizing in 16-bit per color plane comes from
a) reducing the potentially improved supporting electronics noise,
b) less quantization noise and read noise,
c) more accurate demosaicing in the absence of noise
d) no need to amplify the 14-bit signals to 16 bit for postprocessing, so less posterization risk.
The IQ3 100 files in 16-bit mode should be very robust in postprocessing, more than those from a 14-bit origin (even on the IQ3 100).
The improved dynamic range in 16-bit mode is measurable and in actual practice exceeds what a 14-bit capture could even theoretically achieve.
How visible/'tangible' those differences are, obviously also depends on the subject matter, and how good the Raw conversion can handle the signals.
Cheers,
Bart
-
The Proof/Conclusion: Do I get a better image from the Hassie file than the Sony?
Absolutely!
I have compared hundred of shots from the Sony to my files from the H5D 50c WiFi I had and there is not contest.
I have tweaked the Sony file every way. Can't get it the way I need.
But only in color depth! Resolution/sharpness excellent.
The 3FR file with it's 16 bits (regardless of how accomplished) is what gives me the smooth transitions that I want for how I push an image to get the black and white zones the way I want them.
Just can't do it with the Sony, but I do admit, I really wish I could!!
Now, do I need the attributes of the 3FR file for a possible career in Architectural work?
Not really. But I am finding that folks give their time away doing that work OR there are too many already doing a great job in that market here to spend my time breaking in.
I love landscapes, clouds, sea shots. Will it make me serious money, probably not. Does it make me Happy, You can bet on that!
For my passion 3FR works.
Jack
-
I meant that the GFX with their 14bit file gives me the same quality my IQ350 ("16bit" files) gave me. My IQ180 gives me more and my IQ180 gives me less.
Even though that my d800/810 had more dynamic range, I preferred the tones and colours of my "old" Iq180.
I hope that explains it a lot.
Besides that there are far more knowledgeable people around here who can explain that up until the 100Mp backs there was no native 16bit sensor.
Gesendet von iPhone mit Tapatalk
-
Hi,
One point may be that Phase One has a new 16-bits file format for the IQ3-100MP. The old formats were actually 14-bit, blown up to 16 bit in conversion. I would call that a marketing truth, what would normally be a lye. My understanding is that Hasselblad file format was actually 16 bit, but only carrying 14-bits of data.
Also, Steve Hendrix of Capture Integration has confirmed that the old backs were 14-bits.
So, sixteen bitness of MFD was true marketing BS, until the arrival of the 100 MP backs.
But, very clearly, many MFD users find the MFD files superior compared to say Sony A7rII, how come? Why compare with Sony A7rII? Because it has similar MP and has very good DR. Nikon D810 has fewer MP and Canon 5DsR lacks the DR.
Some possible explanations:
- Different processing engines used. I don't know about Phocus, but Capture One applies both extreme sharpening and extreme noise suppression by default.
- It is feasible to assume that DSLRs are balanced to to be used in mixed/low light conditions. Artificial light is often rather low Kelvin and contains little blue. So a camera built for artificial light may be made sensitive to blue with reduced sensitivity in red. Shoot blue sky with a polariser and you may end up with very little signal in the red channel. I don't think MFD-s are balanced for shooting in candle light, but DSLRs are expected to do it and do it well.
- The larger sensor on he MFDs captures more light, which is beneficial for mid tone signal to noise ratio.
- On the other hand, MFDs and raw processors may tend to underexpose in order to protect highlights. Capture One's default gradation curve brightens the image about one stop. So, if you open the same image in say Lightroom and C1, Lightroom may be just great with no exposure compensation while C1 may need -1 EV or so. This is the case on C1/P45+, that is the only MFDB I own.
So, I feel there are a lot of variables.
Medium format devices can probably offer great image quality. But, the cause of that is probably not 16-bit files, rather:
- An advantage in sensor size that is beneficial with regard to noise and also MTF.
- Lenses may be better in many cases. Less sharpening may mean less noise.
- In the Capture One case we have a combination of extreme sharpening and extreme noise reduction, but C1 is good at suppressing both halo artefacts and aliasing artefacts.
So, there are a lot of variables involved.
Going full frame 645, MFD probably has a significant advantage at a significant cost. With 33x44 mm, I don't know. The new systems X1D and GFX deliver great image quality at a very decent price point. But, a very good workflow is needed to take care of it. A camera like the A7rII offers accurate and relatively efficient AF, sensor based shake reduction and the option to use any lens ever made on this planet.
No doubt, an IQ 3-100MP on a technical camera with a proper sample of a Rodenstock HR lens on a steady tripod will beat an A7rII. But, if you shoot hand held, image stabilisation may be more important than 100 MP combined with 15 EV of DR.
Horses for the courses…
Just to say, my experience with MFD is limited to the P45+ on a Hasselblad V system and checking out a lot of raw files from sites like DPReview, Imaging Resource, Digital Transitions and some files I got from Tim Ashley, Voidshatter and others.
Best regards
Erik
I meant that the GFX with their 14bit file gives me the same quality my IQ350 ("16bit" files) gave me. My IQ180 gives me more and my IQ180 gives me less.
Even though that my d800/810 had more dynamic range, I preferred the tones and colours of my "old" Iq180.
I hope that explains it a lot.
Besides that there are far more knowledgeable people around here who can explain that up until the 100Mp backs there was no native 16bit sensor.
Gesendet von iPhone mit Tapatalk
-
Erik,
Back when I was developing software on the SGI platform, what we could do with our proprietary tools would take a file from the Eikonix and the early Leaf and Hassie digital images (both shot and scanned) and dramatically improve them. If you up res the color bit data intelligently, actually VERY intelligently, you can get an out standing image as long as you work in RAW.
That said, I'd expect how Hassie is writing it's RAW file is doing just that but with more intelligence than Sony. Also, yes the larger the capture well in the array the more color fidelity you will get.
Just my 2 cents.
Jack
-
Hi Jack,
Could be…
It seems that the raw files are a bit cooked on many systems. For instance, it could be that the raw channels are bit scaled/normalised. I see less variation than I would expect to see…
Best regards
Erik
Erik,
...
That said, I'd expect how Hassie is writing it's RAW file is doing just that but with more intelligence than Sony. Also, yes the larger the capture well in the array the more color fidelity you will get.
Just my 2 cents.
Jack
-
It seems that the raw files are a bit cooked on many systems. For instance, it could be that the raw channels are bit scaled/normalised. I see less variation than I would expect to see…
In the case of the GFX, I see no evidence of cooking. CDS is not cooking, right?
Jim
-
Hi Jim,
No, CDS is not cooking. But, I think that RGB channels may have different scaling into raw.
Another observation, if you look at a Hassy raw file, it is clearly 14 bits, as raw data has spikes with interval of four, but there is a continuum of data in all channels. That indicates 14 bit of data with some processing applied.
So, I think that raw data is not really raw, it may be modified enough to hide some of the basic differences like spectral sensivity of different pixels.
Best regards
Erik
In the case of the GFX, I see no evidence of cooking. CDS is not cooking, right?
Jim
-
So have just stumbled across this thread again and can confirm that there are some extremely knowledgeable people on here (and that does not include me ;)
So jumping ahead to guessing on the future Sony 33x44mm sensor 100mp sensor, do you guys think that this will contain 16 bits within it?
Thanks in advance!
-
So have just stumbled across this thread again and can confirm that there are some extremely knowledgeable people on here (and that does not include me ;)
So jumping ahead to guessing on the future Sony 33x44mm sensor 100mp sensor, do you guys think that this will contain 16 bits within it?
If Sony can get the read noise down far enough, then I think we will see more precision. I expect that that they will add another bit to the ramp ADCs. Making them 16-bit would probably slow down the readout unacceptably, but they took that approach with the big 100MP chip, so they might do that again. Regardless of the actual conversion precision (15 vs 16 bits), they will probably chunk the data to 16 bits in uncompressed files -- they already did that with the 14-bit data in the a7RII. If we see a 33x44 100MP 15 or 16-bit sensor in a camera from Sony, they will probably extend the craw compression scheme to adapt to it, but I wish they would just do lossless compression and be done with it.
Jim
-
Does Jim or anyone else have an idea of the full-well capacity of the 50MP 44x33 or the 100MP 52x40mm Sony sensors, or other recent Sony sensors?
My guess is that Sony will stay at 16-bit output in all future DMF sensors, but I wonder if the 3.8 micron pixel pitch (14.5 square micron photosites) of 100MP in 44x33mm sensor will benefit form more than 14-bits, even with neglibible read noise. How many bits are needed to report the exact count of every electron in a full well? (14 bits covers 16,384 electrons; 15 bits would cover up to 32,768, and so on.)
-
Does Jim or anyone else have an idea of the full-well capacity of the 50MP 44x33 or the 100MP 52x40mm Sony sensors, or other recent Sony sensors?
My guess is that Sony will stay at 16-bit output in all future DMF sensors, but I wonder if the 3.8 micron pixel pitch (14.5 square micron photosites) of 100MP in 44x33mm sensor will benefit form more than 14-bits, even with negligible read noise. How many bits are needed to report the exact count of every electron in a full well? (14 bits covers 16,384 electrons; 15 bits would cover up to 32,768, and so on.)
FWC of the current GFX sensor is about 50000 electrons. The RN is the key determinator of the necessary precision until it drops to the point where you close to counting electrons, call that half an electron rms at base ISO. When we start getting to those kinds of noise levels, I get a little out of my depth. Eric Fossum is the expert there (and pretty much everywhere else, when it somes to CMOS sensors).
Jim
-
FWC of the current GFX sensor is about 50000 electrons. The RN is the key determinator of the necessary precision until it drops to the point where you close to counting electrons, call that half an electron rms at base ISO.
Thanks Jim. Is there a convenient reference for data like FWC and the base noise level? Is there any such data for recent Sony BSI sensors of similar pixel pitch (about 3.8 microns) Fort this, I would ignore the unusual sensor in the new Sony A9, which reportedly has relatively high read noise and low DR at base ISO speed as the price of its very high speed read-out, according to DPReview at https://www.dpreview.com/articles/7266455439/sony-a9-real-world-iso-invariance-and-dynamic-range