[TonyW]

Comparing the 12 bit Ricoh GR to the now 14 bit GRiii shows me how the DR changes with bit depth in reality. A 16 bit architecture would be a step change in FF cameras.

As we have linear captures it is easy to illustrate (see attachment) what happens with bit depth and DR and how increasing bit depth affects our overall and usable DR.

Your close figures for the 645z compared with the K1ii will raise eyebrows amongst the 645z users.

Why should it?

I've not handled the 645z in the field, but my limited test comparisons between it and a K1 have not shown any significant DR improvement in many use cases.

That is correct the difference between the DR figures of 11.60 and 11.70 is insignificant and can probably be safely discarded as having any real effect. Similarly my own test of 12 stop DR for the 645z (calibration chart shown in attachment) is not that significantly different and would only be relevant to compare ranges if exactly the same tests using the same equipment and conditions were undertaken with a different camera model

Migrating to 16 bit would create much more usable and manipulatable data. This would be nice in the future and would incentivise me to spend rather than greater MPs.

I agree it would be nice to have but I suspect it will be a long wait to see if it trickles down from the top end cameras

Attached Images

   

[normhead]

So what empirical evidence is there that these differences are significant to the human eye?

if there are areas significance, what are the parameters that would define them?

The obvious criticism of this type of graph is that 8 bit images can be quite acceptable. Dynamic range is more determined by the DR capability of the sensor than the number of bits in capture. My ZS100 will never have the DR of my K-1 and K-3 no matter how many bits the output is.

Looking at the top graph, the obvious (to me) question is how does a K-1 achieve a measured 14.9 stop DR on a 14 bit sensor? The hypothesis is disproved by actual scientific measurements.

It all depends on this.

Migrating to 16 bit would create much more usable and manipulatable data. This would be nice in the future and would incentivise me to spend rather than greater MPs.

That's a pretty bold hypothesis. It will definitely need an empirical evidence before I buy in to it.

Not everything that can be graphed is graphing useful significant data.

14 bit (36 MP)


12 bit (12 MP)


I'm going to venture that there is so little difference between 12 bit and 14 bit it makes absolutely no difference to anything.
And the corollary to that will be , going from 14 bit-16 bit will provide even less benefit I any.

The difference in these use shots is more about 12 MP vs 36 MP. Any DR differences are minor.

But I'm open to any real world world proof of concept photos anyone cares to share.

The assumption that camera makers didn't decide 12 bit was suficient, but went to 14 bit as a bit of redundant capacity, making 16 bit just more redundant capacity, is unwarranted.

The notion that 16 bit would provide DR the sensor doesn't support, is just silly.

And the benefits of 16 bit images are unsupported by empirical data.

The "more is always better" folks seem to think their reference for more of whatever somehow translates into better. Even too much water and there things critical to life can kill you. More is not always better. That needs to be established with sm kind o data. Speculative graphs based on unknown data samples (assuming there were any data) and this graph isn't just a hypothesis) don't count as data. Real world applications count as data. You can't make an image with a graph. We need t be able to evaluate the rigour of the graph maker.

Bottom line, once optimal is achieved, "improvements" after that are by definition less than optimal. I see little evidence that 12-14 bit isn't optimal. But hey convince me with images. I'll be happy to change my mind.

What I'm saying is it looks like a bunch of pseudo scientific gobbledygook to me.

What seriously needs to be addressed by the "more is always better crowd, is the hypothesis that competing systems, AD conversion, bit depth , human perception etc will lead to an optimal convergence, and that going above or below those setting produces a decrease in performanence... when you something like proposing 16 bit , bit depth, you really have to demonstrate in the real world, that's even useful, forget about optimal. Addin another 5 horse power to a 395 horse power engine does practically nothin g int erms of power prod but 400 sound a lot better that 395 to a possible purchaser. Not all "bigger is better" is about performance.

My guess is those currently selling 16 bit cameras are using it as marketing hook to justify overpriced product.

[ncallender]

The raw output of a sensor is linear while we then convert that data into curve in some way. That is how the k1 14 bit sensor can achieve a visual 14+ dynamic range. Sensor bit depth doesn't translate visually.

That's not to say that I don't think a 16bit output on full frame would be useless. I love that along with a sensor that could get 16-18 stops of dynamic range and I think we would never need graduated filters any more.

[normhead]

The raw output of a sensor is linear while we then convert that data into curve in some way. That is how the k1 14 bit sensor can achieve a visual 14+ dynamic range. Sensor bit depth doesn't translate visually.

Exactly, it has little to do with the number of bits recorded. It's an output feature.
Where as a sensor with more sensitive pixel sites and deeper wells that would have more DR in it's input, could be translated to 8,12,14 or possibly even 16 bit depending on application demands.

[biz-engineer]

So what empirical evidence is there that these differences are significant to the human eye?

Very good question. Long ago I had a discussion with engineers developing the electronics for large plasma TV (was trendy at the time). They were saying that their minimum color depth specification should be 6bits effective per channel (64 levels per channel) and beyond that most people couldn't tell the difference, so they were doing 8bits video devices to match that spec.12 bits, 14bits, 16bits are there for allowing higher ISO (digital), or headroom for image processing. No one can tell the difference between 12bit and 16bits at base ISO.

---------- Post added 06-03-21 at 18:15 ----------

That is how the k1 14 bit sensor can achieve a visual 14+ dynamic range.

K1 pixel level DR is less than 14ev. The 14+ dynamic range figure is from a calculation based on downsizing images to 8Mpixels.

[BarryE]

Actually, am I correct in thinking that often (K1) doesn't actually use the full 14bit available, more like 13bit plus (maybe one for error correction?) ?

Extra DR, to me, is in the processing scope and the tolerance in capturing images in woodland where the sky is filtering through the tree tops and exposing for the trees, blows the sky. Bracketing can help, but when the sky is in many patches, this is not possible. Neither are filters. Yes, aesthetical, we can accept blown sky patches, but as an example that happens regularly, if extra bit depth bought a decent say a couple of stops I'd go for it. I shoot a lot in the woods.

For controlled shooting, studio, average conditions, extra DR is irrelevant.

[swanlefitte]

This guy did his own testing for himself. He found the software matters as well. His results summed up to this. "Ultimately, I find that at ISO 64 with a Nikon D850:

12-bit files can be pushed 3-4 stops in LR or CO
14-bit files can be pushed 5-6 stops in LR or 4-5 stops in CO
As I generally try to avoid more than 3 stops of shadow recovery due to noise, the color cast due to 12-bit files is rarely going to be an issue in my work. "
8, 12, 14 vs 16-Bit Depth: What Do You Really Need?!
He shows his work so it seems relevent to this thread.

[BarryE]

This guy did his own testing for himself. He found the software matters as well. His results summed up to this. "Ultimately, I find that at ISO 64 with a Nikon D850:

12-bit files can be pushed 3-4 stops in LR or CO
14-bit files can be pushed 5-6 stops in LR or 4-5 stops in CO
As I generally try to avoid more than 3 stops of shadow recovery due to noise, the color cast due to 12-bit files is rarely going to be an issue in my work. "
8, 12, 14 vs 16-Bit Depth: What Do You Really Need?!
He shows his work so it seems relevent to this thread.

Well a 16-bit workflow discussion is going even further off the thread. But, it makes sense to keep in 16-bit as long as possible and switch to 8-bit at the last point, if required, for output. Starting off with full 16-bit data from the camera would further increase the scope for manipulation, if required - many shots, of course, don't need this range, but sadly a few do.

[photoptimist]

The biggest value of bit depth (assuming it's backed by a sensor with sufficient DR) is in the expansion of the usable exposure triangle.

Although people have reported K-1ii files are easier to post process, and there was no change in bit depth. How can you be sure it was bit depth that made the difference? Even if something changed, attributing it to bit depth may be erroneous.

You are somewhat correct here. Just slapping a 16 bit A-to-D convertor into any camera isn't going to improve the results. The sensor pixels need to have a sufficient well depth and the analog electronics need to be clean enough to output something like 16-bit DR.

As well there is the law of diminishing returns, as pointed out recently discussing the lack of improvement in DxO scores going to 50 MP on FF. It may well be that 12 bit depth was optimum, 14 bit depth is overkill, and 16 bit would be un-noticeable. I'm not sure how much you can lean about a specific attribute of a camera like bit depth, when you don't know what else was improved.

16 bits DR adds two more stops of usable shutter speed and aperture in Tv, AV, TAv modes.

It may be true that shooters who always have lots of light, use shallow DoF, and do tripod work with stationary subjects will always have exposures at near base ISO and will see only minuscule differences in the darkest shadows between 12, 14, and 16 bits. They'll only get value from 16-bits in HDR-like situations (e.g., a shot of the interior of a house that also include a view of the sunlit garden outside). Another classic scenario that benefits from more bits is scenes with holiday lights -- the exposure needs to set to avoid clipping the lights (so the color of the light is accurately recorded) but the objects lit by those lights will be extremely dark due to the r-squared fall-off of intensity from the tiny bulbs to the distant surfaces.

But those who lack light, shoot for deeper DoF, and have fast moving subjects will strongly benefit from being able to use 2 extra stops of ISO to enable an extra stop or two of faster shutter or deeper DoF. That's been my empirical experience -- every added stop of usable high ISO performance expands my opportunities to shoot macro in darker forests, catch faint nebulae in astrophotography, get decent shots in dimly-lit museums and cathedrals, get good shots around the campfires, etc.


For me, cameras are no where near the point of diminishing returns where bit depth is concerned. And more importantly, the law of diminishing returns is faulty logic where cameras are concerned. Even if a niche feature only applies to 1% of shooting situations, if it lets you get images you could not get before, it's a new realm for creativity and subject matter that wasn't accessible before. Having 16-bit is like having 1/8000 sec shutter speed -- both are past the point of diminishing returns and yet they are also essential to expanding entire genres of photography.

[TonyW]

I have to say that I am having a hard time trying to read into this post as it appears to be a rant against knowledge. But I could be wrong and just maybe it is actually a search for knowledge and understanding. My assumption is also that as graphs are mentioned the poster has an issue with me and my data. So I will try my best to explain the facts of digital capture where appropriate

So what empirical evidence is there that these differences are significant to the human eye?

if there are areas significance, what are the parameters that would define them?

Relating to what exactly?

The obvious criticism of this type of graph is that 8 bit images can be quite acceptable. Dynamic range is more determined by the DR capability of the sensor than the number of bits in capture. My ZS100 will never have the DR of my K-1 and K-3 no matter how many bits the output is.

No one has stated that 8 bit images are not acceptable so why bring it up as it is not part of this discussion?

Dynamic range is determined by the DR capability of the sensor in that you are correct. Your sensor is an analogue capture device (counts photons) but that data needs to be sampled and that is via the A/D converter which is a digital process (which discards some data along the way). Linear A/D converters will convert exactly the way that is shown in the spreadsheet tables and will limit the Dynamic range as shown - this is quantifiable scientific fact!
Bear in mind that the values shown are theoretical maximum levels for each camera 8 to 16 bit - your camera may not actually reach the maximum but should not be too far out

I do not know what bit depth your Z3100 or K3 claims but its DR will be limited to its A/D converter and if that is 14 bit then the results should be pretty close

EDIT: In an effort to help you see the limitations (not that this is that important)
https://www.photonstophotos.net/Charts/PDR.htm#Panasonic%20Lumix%20DMC-ZS100...1,Pentax%20K-3

Looking at the top graph, the obvious (to me) question is how does a K-1 achieve a measured 14.9 stop DR on a 14 bit sensor? The hypothesis is disproved by actual scientific measurements.

It seems we must be looking at different graphs!! My table only shows 8 - 16 bit sensor values and no mention of K1, and where on earth did the 14.9 stop DR come from? Did you make the hypothesis because I certainly did not. And if you are saying that scientific measurements proved that it is not possible to get more than 14 stops DR from a 14 bit capture then we are in agreement.
If a 14 bit system is seen to be capturing more than 14 bits then I would suggest one of two possibilities. 1. The data is suspect, 2. The A/D converter is actually capturing more bits than declared by the manufacturer - the question would then by why is the manufacturer hiding this.

Not everything that can be graphed is graphing useful significant data.

True, but everything that has been graphed here by me is significant and true and can be objectively proved!

14 bit (36 MP)

12 bit (12 MP)

Nice images but of no relevance to output of 8,12, 14 or 16 bit discussion

I'm going to venture that there is so little difference between 12 bit and 14 bit it makes absolutely no difference to anything.
And the corollary to that will be , going from 14 bit-16 bit will provide even less benefit I any.

Venture all you want but this is only your subjective view.

The assumption that camera makers didn't decide 12 bit was suficient, but went to 14 bit as a bit of redundant capacity, making 16 bit just more redundant capacity, is unwarranted.

The notion that 16 bit would provide DR the sensor doesn't support, is just silly.

The notion that 16 bit A/D converter would be added by a manufacturer to a sensor that is not capable of counting the photons is nonsensical.

The "more is always better" folks seem to think their reference for more of whatever somehow translates into better. Even too much water and there things critical to life can kill you. More is not always better. That needs to be established with sm kind o data. Speculative graphs based on unknown data samples (assuming there were any data) and this graph isn't just a hypothesis) don't count as data. Real world applications count as data. You can't make an image with a graph. We need t be able to evaluate the rigour of the graph maker.

Speculative graphs!! Well I do not know what you have been looking at but if you are referring to mine you clearly have a problem with understanding. If this is my fault and I have not made it clear then ask away!

What I'm saying is it looks like a bunch of pseudo scientific gobbledygook to me.

What you are saying apart from being rude is nonsense and comes from what appears to be a lack of understanding.

To be as clear as I can be on the tables and graphs I showed. The first table illustrates the DR potential for 8, 12, 14 and 16 bit captures - it is based on the way that digital image acquisition works and its limitation. The second so called 'Graph' is the measure of a Pentax 645z capability relating to DR and is a real world calibration of a Sekonic L758DR spot meter i.e. it is an applied profile to the meter to mimic how the camera responds

[texandrews]

OK, so once again we are veering into incivility territory.


Extraordinary, this seems to be popping up again and again. Yesterday a thread got closed for just this reason, and the mods have all been busy with warnings, infractions, deleted posts, trimmed posts, & etc.


So, the choice is yours: Close this one as well? Or moderate the rhetoric? It's an easy to follow track---don't write anything that gives offense, don't rise to a perceived one.

[TonyW]

The raw output of a sensor is linear while we then convert that data into curve in some way. That is how the k1 14 bit sensor can achieve a visual 14+ dynamic range. Sensor bit depth doesn't translate visually.

That's not to say that I don't think a 16bit output on full frame would be useless. I love that along with a sensor that could get 16-18 stops of dynamic range and I think we would never need graduated filters any more.

Most digital cameras use a 10 -14 bit A/D converter so their theoretical max. DR is 10 -14 stops. The higher bit depth helps minimise posterisation in editing. DR is usually limited by noise levels (and your acceptance of noise). Even a high precision A/D converter will not record a greater dynamic range and in practive the DR of a camera may not even approach the A/D converter theoretical max. For a 10 - 14 bit conversion you can generally only expect 8-12 stops usable range. While the raw sensor output is linear and has to be converted to a TRC you will not get more DR as the limits are already baked into the raw data and therefore DR is fixed

[Michael Piziak]

Smartphones usually have QuadBayer arrangements, so the MP in the final picture is much lower than the nominal (advertised) value.

Regarding MP count at large, I find that there has to be a balance between camera MP count, interface/card/mass storage transfer speed, HDD size and cost, CPU processing speed/RAM cost/performance and software speed in general.
All or most of these factors need to be aligned at reasonable values, otherwise the whole Jenga tower collapses.


With 8TB HDD being cheap-ish, PCs being quite powerful, SD cards being big and cheap, many manufacturers implementing USB 3 in their cameras etc. I'm starting to feel like I manage with more than 16 MP per shot 24-36 looks like my personal sweetspot ATM, but it's largely a personal preference.
After all, if one doesn't print, you can't really have more than ~8 MP on screen at a time

For professionals it's another story entirely, obviously.

I just spent over an hour reading about how ARM chip technology is going to completely speed up and change the computer laptop, desktop, and server world. And this is happening before our very eyes. There are already demos on youtube where smartphones with ARM are completely out benchmarking loaded desktops.

[normhead]

OK, reading the article posted above....
8, 12, 14 vs 16-Bit Depth: What Do You Really Need?!

In summary:

Do not shoot JPG (8-bits)
A 12-bit RAW file is excellent for most work and offers significant space savings over 14-bit RAW. This is the best choice if you care about file size.
If you want the absolute best quality in the shadows, shoot 14+ bit RAW files (ideally with lossless compression to save space). This is the best choice if you don’t care about larger files and shoot scenes with wide dynamic range (deep shadows).
If you can shoot 16-bits, that’s fine but probably overkill. Worth testing your camera to see if you can use a lesser setting to save on file size.

Pretty much what I said.
Anyone else have some practical stuff, that doesn't agree with my posts, this one was right up my alley.

[Ian Stuart Forsyth]

I agree it would be nice to have but I suspect it will be a long wait to see if it trickles down from the top end cameras

I will also add that how color data is collected mainly the red channel in the majority of the lighting conditions we shoot in, that channel will receive less light and this can be as much as 1- 1.5 stop less than the other 2 channels

This can increase how noise appears, so even in an image that is only using 8 stops of DR we are be pushing the red channel down into 9 and 10 stops from full saturation capacity of the sensor and this can only decrease the quality at which we can reproduce colors in our photographs.