12bit vs would 14bit

I was wondering if in new Ursa 4.6 (and mini) would be possible to use 14bit RAW.

CinemaDNG in BMCC/BMPC/URSA at this point delivers 12bit footage, it’s a bit over 4000 levels/values = data. Arriraw (for example) delivers 14bit raw which is over 16000 levels/values. It's very big difference. Basically, 14 bit raw contains four times more data in each frame.

Would be interesting to know if it possible in this case.

Probably but I doubt they will. 12bit log encoded raw has plenty of bit depth to store the dynamic range for even the 4.6k at about 85% of the file size. After grading the 8bit slog2 from the a7s I believe people put too much importance on bit depth as it can be pushed a lot even at only 8 bit. If anything I would be curious to see a 10 bit raw option, but that is just my opinion.

Pete, remember the BMD cameras are outputting 12bit log, not 12 bit linear; the raw is debayered into 16bit linear in Resolve and 32bit floating point is used to calculate all colour values. The 16bit linear in camera is derived from 22bits of data from the sensor.

Edit: note the comparative information related to ARRI raw below.

Rick Lang
Sent using Tapatalk HD

Hallo,
To clarify, I actually work at ARRI. ARRIRAW is an uncompressed and unencrypted 12-bit Logarithmic RAW format. We take 64 output channels off our ALEV III sensor (x2 pairs of 32, one high gain, one low gain) which are delivered to the 14-bit A/D converters which can reconstruct 16 bit luminance information of the sensor into the 12-bit logarithmic format.

http://www.arri.com/camera/alexa/techno ... as_sensor/

http://www.arri.com/camera/alexa/learn/arriraw_faq/

Chase, thanks for the post!


Rick Lang
Sent using Tapatalk HD

You are very welcome Rick!

To answer Mr. PetePolyakov, I personally believe that in order to achieve the most benefit from 14-bit RAW files that the sensor would have to have an exceeding large number of output paths and the A/D converter would ideally have 16-bit precision (16-bit A/D) to convert the data to a 14-bit logarithmic data (14-bit linear contains a similar amount of data/information as 12-bit logarithmic). The processing power of the A/D converter would most likely come at an extreme price cost as the only 16-bit A/D converter equipped camera is the Sony F65 to my knowledge (RED does not provide this information). Also 14-bit RAW data would need considerable storage bandwidth and space compared to 12-bit especially at higher frame rates and would come at quite a cost.

Hallo,
To clarify, I actually work at ARRI. ARRIRAW is an uncompressed and unencrypted 12-bit Logarithmic RAW format. We take 64 output channels off our ALEV III sensor (x2 pairs of 32, one high gain, one low gain) which are delivered to the 14-bit A/D converters which can reconstruct 16 bit luminance information of the sensor into the 12-bit logarithmic format.

http://www.arri.com/camera/alexa/techno ... as_sensor/

http://www.arri.com/camera/alexa/learn/arriraw_faq/

The enemy is among us!! jk

Thanks for sharing this information Chase! It's great to learn these details from the horse's mouth, so to speak.

Thanks, the free exchange of information is what this site is supposed to be about. Excellent info Chase thanks again.

Hallo,
To clarify, I actually work at ARRI. ARRIRAW is an uncompressed and unencrypted 12-bit Logarithmic RAW format. We take 64 output channels off our ALEV III sensor (x2 pairs of 32, one high gain, one low gain) which are delivered to the 14-bit A/D converters which can reconstruct 16 bit luminance information of the sensor into the 12-bit logarithmic format.

http://www.arri.com/camera/alexa/techno ... as_sensor/

http://www.arri.com/camera/alexa/learn/arriraw_faq/

Posts like this make me wish once again for a "like" feature on the forums. There's a report button, but no like button. This post is a good reason to add a like button.

Ah you are all to nice! I am only happy to clarify some minor information! Danke!

Chase, I'm curious and hope don't sound ridiculous and further that I'm not stepping to far... Not being as technical minded as some here how would you characterize the proposed sensor the the Ursa Mini?

Probably but I doubt they will. 12bit log encoded raw has plenty of bit depth to store the dynamic range for even the 4.6k at about 85% of the file size.

Sorry to be pedantic, but any bit depth can "store" any dynamic range, as bit depth and dynamic range are two independent and unrelated properties.

Probably but I doubt they will. 12bit log encoded raw has plenty of bit depth to store the dynamic range for even the 4.6k at about 85% of the file size.

Sorry to be pedantic, but any bit depth can "store" any dynamic range, as bit depth and dynamic range are two independent and unrelated properties.

I doubt anyone here would agree that a 1 bit file can faithfuly store the dynamic range of these cameras which is what this thread is about.

Probably but I doubt they will. 12bit log encoded raw has plenty of bit depth to store the dynamic range for even the 4.6k at about 85% of the file size.

Sorry to be pedantic, but any bit depth can "store" any dynamic range, as bit depth and dynamic range are two independent and unrelated properties.

I doubt anyone here would agree that a 1 bit file can faithfuly store the dynamic range of these cameras which is what this thread is about.

Well you just made his point for him Black and white, aka 1bit, covers 100% of any cameras dynamic range, from the blackest black, to the whitest white the camera can record.

You are talking about gamma graduation, or steps in between black and white clip, which requires bit depth.

I doubt anyone here would agree that a 1 bit file can faithfuly store the dynamic range of these cameras which is what this thread is about.

Well you just made his point for him Black and white, aka 1bit, covers 100% of any cameras dynamic range, from the blackest black, to the whitest white the camera can record.

You are talking about gamma graduation, or steps in between black and white clip, which requires bit depth.

A one-bit pixel can certainly render the whitest white and the blackest black. However, a better illustration of my point using a one-bit system comes from the fact that one-bit images can show infinite gradation between the whitest white and the blackest black.

In truth, for over 100 years we have been using one-bit images that show gradation. You can see such images everywhere -- in magazines and newspapers, on posters and billboards, in brochures and flyers, etc. These images are screen printed, and such images are comprised of a matrix of black dots on a white field (let's restrict ourselves to black & white images, for simplicity's sake). So, these images actually are made from only two shades -- or "one-bit."

Of course, screen printing does have an analog component relating to the size of the black dots. However, those various sized black dots can easily be rendered digitally using a digital, one-bit system of significantly higher resolution, in which each screen printed dot is made up of numerous black or white digital pixels.

The next point that usually appears in discussions such as this relates to the notion that a one-bit camera/sensor could not work. However, a one-bit camera/sensor that captures gradation is absolutely possible merely by varying pixel-specific filters/apertures to give adjacent pixels various sensitivities. Varied pixel sensitivities has already been demonstrated by Panavision with the design of their abandoned Dynamax sensor, and varied pixel sensitivity can also be done electronically, similar to Magic Lantern's "dual iso" technique.

Go to a museum gift shop and look at their finest, glossy black & white photography books/magazines. There is a reason why these essentially one-bit images can exhibit such an amazing photographic range of tones. It is because bit depth is just one of two primary factors in color depth (or, in our case, shade depth) -- the other major (and equally important) factor of color depth is resolution.

I will also repeat my above point, which has absolutely no relation to the rest this post -- bit depth and dynamic range are two independent, unrelated properties (which can be easily demonstrated).

Mr. Harrington, As I can not comment on any camera sensor that has not been tested by respected sources I will only offer this as an experience I had, At NAB I have asked the BMD Camera Product Manager (I was directed to him by a friendly BMD associate) about how the 15 stops of Dynamic Range was measured of this new sensor and at what ISO this was measured but he was unable to provide me with these answers. As you may have read that at ARRI we measure our cameras dynamic range using an ARRI DRTC-1 (Dynamic Range Test Chart) which is a widely respected professional non-propriearty chart that can be used and purchased to test the dynamic range of a camera system. You can look up how this chart works with the attached link.

http://www.arri.com/arriajax?mod=produc ... roduct=263

Any sensor that can offer 15 stops of latitude at a respectable ISO (800 perhaps) would be most innovative if it lives up to its marking hype.

Chase, you can be sure when the production camera ships, it will be measured and poured over by those interested in the rolling shutter and global shutter dynamic range at all the ISO ratings available. This was done for the well regarded BMCC and that sensor was just 'very good.' This 4.6K sensor might touch on 'extraordinary' if it does appear to perform anywhere near as advertised.


Rick Lang
Sent using Tapatalk HD

Please stop me when I go wrong, because I'm trying to learn here!

bit depth and dynamic range are two independent, unrelated properties (which can be easily demonstrated).

Are you sure camera sensor engineers view dynamic range and bit depth as unrelated? This to me indicates that they are related (in a current camera sensor, not in print on paper):
http://en.wikipedia.org/wiki/Dynamic_range
http://en.wikipedia.org/wiki/Analog-to-digital_converter

Black and white, aka 1bit, covers 100% of any cameras dynamic range, from the blackest black, to the whitest white the camera can record.You are talking about gamma graduation, or steps in between black and white clip, which requires bit depth.

Once again, I don't know how sensor engineers define "dynamic range", but the word certainly often is used with the same meaning as "contrast ratio". But why coin the phrase "dynamic range" if it means the same as "contrast ratio"? The general meaning of the words "dynamic" and "range" to me, at least, indicates that "dynamic range" is something that continually changes between a series of values. And "stops", the measurement used for "dynamic range", indicates that the value doubles a certain number of times. It doesn't just go from one value (white) to another (black) and back again. Shouldn't, therefore, "dynamic range" be something that doubles a certain number of times, 12 for instance, and incorporates those 4096 discrete values?

You might say that all cameras can record any luminance value; it's the luminance level at which they turn everything brighter to whitest white that differs. It's the discrete values that are added that are of interest, not just the luminance at the clip level.

Mâns, one way to look at the dynamic range of a digital sensor is a comparison of the maximum well charge of a photosite and the background noise level of a cell. The number of stops is the log base 2 value dividing the maximum charge by the minimum charge or noise, for example log base 2 (40,000 e- / 2 e-) = log base 2 (20,000) = 14.3. Now we don't know the values for the 4.6K sensor that BMD has designed; we are told that the dynamic range is 15 stops, so they may have a higher total well charge and/or a lower noise level. For BMD's exclusive 4.6K sensor, the charge values may be a well charge of 40,000 but much lower noise due to the cooling as illustrated here: log base 2 (40000 / 1.25) = log base 2 (32,000 = 15!

People often refer to the maximum perceived latitude in a sensor. This can be less than the mathematically determined dynamic range but it's a subjective valuation wherein one person might say they only can see 12 stops and someone else might say 11.5 stops.

Rick Lang
Sent using Tapatalk HD

Thanks Chase, your reply is spot on. Really appreciate that you've taken the time to stop by here. I hope that you stick around and comment from time to time. I'm learning from your posts and as 'they' say 'every little bit helps'. I just keep wondering who the heck 'they' are?

The accuracy of the sensor matters too. If there is a huge amount of noise, or if the sensor is not very efficient, it won't matter what bit depth the A/D works at. Let's say for any given pixel at any given frame the noise level is +/-1%. That would make the extra bit depth nothing but storage waste, as anything beyond ~9 bit would be pointless. Of course none of the sensors in BMD's cameras have a 1% random noise level, but as you increase the precision of the A/D you must also increase the accuracy of the input to make it worthwhile.

I doubt anyone here would agree that a 1 bit file can faithfuly store the dynamic range of these cameras which is what this thread is about.

Well you just made his point for him Black and white, aka 1bit, covers 100% of any cameras dynamic range, from the blackest black, to the whitest white the camera can record.

You are talking about gamma graduation, or steps in between black and white clip, which requires bit depth.

A one-bit pixel can certainly render the whitest white and the blackest black. However, a better illustration of my point using a one-bit system comes from the fact that one-bit images can show infinite gradation between the whitest white and the blackest black.

In truth, for over 100 years we have been using one-bit images that show gradation. You can see such images everywhere -- in magazines and newspapers, on posters and billboards, in brochures and flyers, etc. These images are screen printed, and such images are comprised of a matrix of black dots on a white field (let's restrict ourselves to black & white images, for simplicity's sake). So, these images actually are made from only two shades -- or "one-bit."

Of course, screen printing does have an analog component relating to the size of the black dots. However, those various sized black dots can easily be rendered digitally using a digital, one-bit system of significantly higher resolution, in which each screen printed dot is made up of numerous black or white digital pixels.

The next point that usually appears in discussions such as this relates to the notion that a one-bit camera/sensor could not work. However, a one-bit camera/sensor that captures gradation is absolutely possible merely by varying pixel-specific filters/apertures to give adjacent pixels various sensitivities. Varied pixel sensitivities has already been demonstrated by Panavision with the design of their abandoned Dynamax sensor, and varied pixel sensitivity can also be done electronically, similar to Magic Lantern's "dual iso" technique.

Go to a museum gift shop and look at their finest, glossy black & white photography books/magazines. There is a reason why these essentially one-bit images can exhibit such an amazing photographic range of tones. It is because bit depth is just one of two primary factors in color depth (or, in our case, shade depth) -- the other major (and equally important) factor of color depth is resolution.

I will also repeat my above point, which has absolutely no relation to the rest this post -- bit depth and dynamic range are two independent, unrelated properties (which can be easily demonstrated).

A 1 bit file cannot store infinite levels of gradients as you are describing. What you are talking about is a percieved optical illusion of a gradient or an undersampling artifact. It is an effect that is created by the image and by the observer that are not mutially exclusive. Look at a newspapper under a maginifying glass and those gradients will dissappear.

Are you sure camera sensor engineers view dynamic range and bit depth as unrelated?

I can't speak for camera sensor engineers.

Nevertheless, dynamic range and bit depth are two independent and unrelated properties.

I do not want to keep derailing this thread, so I will try to be concise.

First of all, bit depth is a property that applies only to digital systems while dynamic range applies to both digital and analog systems. So, in the analog world, there is obviously no relation between dynamic range and bit depth, as bit depth does not even exist.

In digital systems, bit depth is a property that merely denotes the number of digital increments mapped to the range between "0" amplitude and maximum amplitude (the dynamic range exists within this minimum-maximum amplitude range). Any bit depth can be mapped within that amplitude range, and, indeed, some digital cameras offer different bit depths with the same dynamic range.

Here is one such camera that offers a choice of 8-bit, 10-bit or 12-bit. Of course, no matter which bit depth one chooses, the camera's dynamic range doesn't change.

A 1 bit file cannot store infinite levels of gradients as you are describing.

Of course, a 1-bit file can store infinite gradients (or "essentially" infinite gradients, as the size of a file is always finite). I gave a simple example above of how it works.

Furthermore, digital files are just digital information -- the same information contained in a 32-bit file can be converted into a 1-bit file, and converted back-and-forth all day with no loss in quality.

What you are talking about is a percieved optical illusion of a gradient or an undersampling artifact. It is an effect that is created by the image and by the observer that are not mutially exclusive. Look at a newspapper under a maginifying glass and those gradients will dissappear.

The exact same "optical illusion" happens when one views a digital image on an RGB monitor -- the observer sees millions of colors with the naked eye, but when one uses a magifying glass, one only sees red, blue or green pixels.

In both the 1-bit scenario and any 8-bit/16-bit/32-bit example, digital information is displayed and the viewer's eye blends the pixels.

If we continue this discussion, perhaps it would be best to start a new thread.

You are precisely correct Chad the sensor output accuracy is quite important but because this is not a concrete measurement taken by all manufactures unfortunately, in a sense that many manufactures do not release this specs in terms of sensor out output paths and noise levels (which can vary with sensor temperature) for example as is discussed in the output paths of the ARRI ALEV III sensor on the ARRI website, it makes it very difficult to verify this sensor noise information from all manufacturers.

You are very welcome Mr. Harrington, happy to be of help!

I can't speak for camera sensor engineers. Nevertheless, dynamic range and bit depth are two independent and unrelated properties. I do not want to keep derailing this thread, so I will try to be concise. First of all, bit depth is a property that applies only to digital systems while dynamic range applies to both digital and analog systems. So, in the analog world, there is obviously no relation between dynamic range and bit depth, as bit depth does not even exist.

But we are talking about camera sensors here. "Dynamic range" means different things in different contexts.

In digital systems, bit depth is a property that merely denotes the number of digital increments mapped to the range between "0" amplitude and maximum amplitude (the dynamic range exists within this minimum-maximum amplitude range). Any bit depth can be mapped within that amplitude range, and, indeed, some digital cameras offer different bit depths with the same dynamic range. Here is one such camera that offers a choice of 8-bit, 10-bit or 12-bit. Of course, no matter which bit depth one chooses, the camera's dynamic range doesn't change.

The question then: What we talk about when we talk about dynamic range. See my previous post:

Once again, I don't know how sensor engineers define "dynamic range", but the word certainly often is used with the same meaning as "contrast ratio". But why coin the phrase "dynamic range" if it means the same as "contrast ratio"? The general meaning of the words "dynamic" and "range" to me, at least, indicates that "dynamic range" is something that continually changes between a series of values. And "stops", the measurement used for "dynamic range", indicates that the value doubles a certain number of times. It doesn't just go from one value (white) to another (black) and back again. Shouldn't, therefore, "dynamic range" be something that doubles a certain number of times, 12 for instance, and incorporates those 4096 discrete values? You might say that all cameras can record any luminance value; it's the luminance level at which they turn everything brighter to whitest white that differs. It's the discrete values that are added that are of interest, not just the luminance at the clip level.

But we are talking about camera sensors here. "Dynamic range" means different things in different contexts.

Not really. Dynamic range is basically the ratio between the strongest signal and the noise floor, in both analog and digital systems.

With transducers (such as camera sensors), dynamic range often refers to the "capture" range.

But we are talking about camera sensors here. "Dynamic range" means different things in different contexts.Not really. Dynamic range is basically the ratio between the strongest signal and the noise floor, in both analog and digital systems.

The minimum level is not always at the noise level. And DR can refer to different things: sound, light, electricity etc.

So, in the analog world, there is obviously no relation between dynamic range and bit depth, as bit depth does not even exist.

But there is in an AD converter.

As I'm not qualified to have an opinion in the matter, I'll just end my part in the conversation by recommending anyone who might be interested to start by reading this:

http://www.amazon.com/VES-Handbook-Visual-Effects-Procedures/dp/0240825187/ref=sr_1_1?s=books&ie=UTF8&qid=1431092454&sr=1-1&keywords=The+VES+Handbook+of+Visual+Effects
http://en.wikipedia.org/wiki/Analog-to-digital_converter
http://en.wikipedia.org/wiki/Signal-to-noise_ratio
http://en.wikipedia.org/wiki/Decibel
http://en.wikipedia.org/wiki/Dynamic_range
http://www.analog.com/library/analogDialogue/archives/39-06/Chapter%208%20Data%20Converter%20ApplicationsF.pdf
http://www.andor.com/learning-academy/dynamic-range-and-full-well-capacity-a-definition-of-ccd-dynamic-range