Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Mysterium-X dynamic range...

By themselves, the numbers are near useless, but compared to another camera, there is a relative difference that the numbers will demonstrate.

I think though that there are two figures that should be released. One is the total measurable dynamic range of both the original sensor and the new MX sensor. The other is the range that begins at the same noise level at the bottom, to show the extra stops that are usable in the new sensor compared to the old sensor.

Because testing may show, let's say (I'm just making up a figure), a one-stop total increase in measurable dynamic range -- but when comparing noise levels, may show two or more stops increase in usable range if you pick the same level of noise as the base.

I'd also like to see the same test in 5600K and 3200K lighting balance.
 
I was skeptical of the Red "look" for a long time. I didn't like it, compared to film. But, these new developments are pretty incredible.

I saw The Lovely Bones recently, and I forgot it was a Red shoot. It looked really good. Was that entirely Red?

(Unfortunately, The Lovely Bones hit numerous story walls and flopped in the end. It should have worked. Oh well.)
 
Joe G. said:
I was skeptical of the Red "look" for a long time. I didn't like it, compared to film. But, these new developments are pretty incredible.

I saw The Lovely Bones recently, and I forgot it was a Red shoot. It looked really good. Was that entirely Red?

(Unfortunately, The Lovely Bones hit numerous story walls and flopped in the end. It should have worked. Oh well.)
Click to expand...

Majority was shot on film http://www.reduser.net/forum/showthread.php?t=15679
 
Chris Clogg said:
Makes me wonder how close 32 bits-per-pixel displays (in the future that is.. currently we only see on 8bpp displays as far as I know) will be able to recreate real/life lighting. No matter how many f-stops our cameras can capture... the displays need to be able to display that range too, otherwise it's just compressed/stylized into 8 bit color like that HDR effect photographers do...

-cc
Click to expand...

Actually, FSI now has a 10-bit grading monitor. They're a superb small company (I'm not affiliated with them--I just own two of their units) making a product that runs rings around the competition in the same price range.
http://www.flandersscientific.com/index/lm2470w
 
Since it was all probably an off-the-clock, over-worked, self-induced, brain-dead, soap-opera: I can't help but be interested in human vision. It is part of the system I am devoted to work for. I don't want to go chat elsewhere, cause I don't want to shoot with anything else. So, if I sound emo, just let me know, clearly, and I promise, I'll stfu. I may be overly-cautious because I realize Jim needs to sell cameras, and forum/www-clueless people may jump in and read a few lines out of context, or put too much weight on any given member's posts. Jim, don't let our/my incessant questioning tire you. We/I ask cause we like. If we didn't like, you'd hear crickets. :appropriate smiley:

In the end I think of Red as a company that got to start wherever they thought was best to pour the initial concrete, without years of pre-conceptions, or heavy overgrown bureaucratic systems associated with large corporations. I know squat about the other big companies so that may be off by a mile, but Red started nimble, and very willing to change. It is obvious, just from looking at a frame off an R1-M, that they have some serious pixel wisdom employed. I wouldn't be in the least bit surprised if they were taking human vision into account when setting their compass, and they sure seem to know winch way is North. :up-beat appropriate smiley:
 
Roberto Lequeux said:
Since it was all probably an off-the-clock, over-worked, self-induced, brain-dead, soap-opera: I can't help but be interested in human vision. It is part of the system I am devoted to work for. I don't want to go chat elsewhere, cause I don't want to shoot with anything else. So, if I sound emo, just let me know, clearly, and I promise, I'll stfu. I may be overly-cautious because I realize Jim needs to sell cameras, and forum/www-clueless people may jump in and read a few lines out of context, or put too much weight on any given member's posts. Jim, don't let our/my incessant questioning tire you. We/I ask cause we like. If we didn't like, you'd hear crickets. :appropriate smiley:

In the end I think of Red as a company that got to start wherever they thought was best to pour the initial concrete, without years of pre-conceptions, or heavy overgrown bureaucratic systems associated with large corporations. I know squat about the other big companies so that may be off by a mile, but Red started nimble, and very willing to change. It is obvious, just from looking at a frame off an R1-M, that they have some serious pixel wisdom employed. I wouldn't be in the least bit surprised if they were taking human vision into account when setting their compass, and they sure seem to know winch way is North. :up-beat appropriate smiley:
Click to expand...

Roberto... you're good. I'm all for the dissecting of the math, science and details. Just wanted to point out that at some point it needs to relate to something in the real world.

The new sensor favorably measures up with anything in the industry. The "I need a Canon to shoot in low light" issue should now be cleared up. I know of no other sensor (that doesn't line-skip) that is as clean. Maybe Arri's will be. I hope so. Then we can talk about other stuff... like frame rates, workflow, and shooting great stuff.

Jim
 
We are finding that those that are shooting the new sensor feel right at home with ISO 800- 2000.

Jim
 
just wondering but.. 12bits redcode = 13stops = how?

..the original epic-x specs. said 12bit sensor, then the stages release says 15bit. However, if the redcode is still 12bit, since the red one is as far as I know, has redcode gone all "log" all of a sudden or have I missed something? sorry if someone else has already queried the math here.. looking forward to all things floating-point :)
 
ronx said:
just wondering but.. 12bits redcode = 13stops = how?

..the original epic-x specs. said 12bit sensor, then the stages release says 15bit. However, if the redcode is still 12bit, since the red one is as far as I know, has redcode gone all "log" all of a sudden or have I missed something? sorry if someone else has already queried the math here.. looking forward to all things floating-point :)
Click to expand...

BITS DO NOT EQUAL STOPS!!! :emote_headwall:

Some of us may have beat the numbers to death previously in this thread, but go back and read the past few pages. Stops are a representation of the range of luminance the sensor can receive and digest. The bits is the precision in which that information is encoded. In this case it's 12 bits. So each pixel on the sensor is encoded with a luma value ranging from 0 to 4095. 12bits = 2^12 = 4096.
 
Jeff Kilgroe said:
BITS DO NOT EQUAL STOPS!!! :emote_headwall:

Some of us may have beat the numbers to death previously in this thread, but go back and read the past few pages. Stops are a representation of the range of luminance the sensor can receive and digest. The bits is the precision in which that information is encoded. In this case it's 12 bits. So each pixel on the sensor is encoded with a luma value ranging from 0 to 4095. 12bits = 2^12 = 4096.
Click to expand...

sure, I understand that, but each stop requires that you double the range of values, so unless you're saying we're getting 13stops from a 12stop raw image, I'm interested just how, mathematically. 13stops would require a range going 0..8192.

Start with 1 and double it.
Each time you do, its a stop. How many do you count before you go beyond 4095? Now if a 15bit sensor is being fed through a 12bit callibrated FLUT, into redcode, that could extend useable range, but obviously by throwing away some less useful signal..

I'm just probing to kill time while the AISCs are born, as excited as everyone else so please don't jump on an honest question
 
ronx said:
sure, I understand that, but each stop requires that you double the range of values, so unless you're saying we're getting 13stops from a 12stop raw image, I'm interested just how, mathematically. 13stops would require a range going 0..8192.
Click to expand...

But a 13 stop image, doesn't require anything. You can represent a 13 stop image with an 8 bit value or even less if you wanted. It's just that as you increase the range, you lose precision. Unless you increase the bit depth.

That said, the Mysterium-X doesn't really add sensitivity or range over the current Mysterium sensor. It increases usable range through a much lower noise floor and better noise handling. So now, it's much the same image as before, but with improved noise handling that allows for usable ISOs of 800-1000+. We are gaining the extra 1.5 stops or so on the bottom end, which is data that before would be crushed out. Additionally, some of the extra range is through the new color science and this will benefit the current RED One Mysterium sensor as well, as people are seeing on the post side of things, but even more so once the rest of it is implemented in-camera.

I'm just probing to kill time while the AISCs are born, as excited as everyone else so please don't jump on an honest question
Click to expand...

I do have to jump on you about the real name thing though. You should stop by the real names thread in Recon and get that taken care of.
 
ronx said:
sure, I understand that, but each stop requires that you double the range of values, so unless you're saying we're getting 13stops from a 12stop raw image, I'm interested just how, mathematically.
Click to expand...
non-linear recording.

Regards, Matt
 
Scott Auerbach said:
Actually, FSI now has a 10-bit grading monitor. They're a superb small company (I'm not affiliated with them--I just own two of their units) making a product that runs rings around the competition in the same price range.
http://www.flandersscientific.com/index/lm2470w
Click to expand...

This is interesting... does the computer using it need a special graphics card? Does the 10-bit color work through DVI?

Is it simpler than I think? E.g. right-click desktop -> Properties -> Display Settings -> and set Color depth to 10 bits? (or I guess it would be 40 bits including RGB + alpha?)

-cc
 
Matthias Hutter said:
non-linear recording.

Regards, Matt
Click to expand...

But it is linear. It's the f-stop scale that is not linear. A sensor doesn't "see" or "understand" stops. It just sees light. Think of a photo receptor on a sensor as a plastic cup, with graduated markings on the side. The scale on the cup would be 0 representing empty, to full at the top represented by a 12bit value (4095).

Now think of the sensor itself as an array of these cups on a football field. And the light that hits your sensor as falling rain. With each read-reset cycle of the sensor, the cups are closed, checked to see how full they are, emptied and then re-opened. The 12bit RAW value for each pixel on your sensor is analogous to how much light (photons) or how much rainfall (raindrops) they collected on that 0 to 4095 scale. It's linear.

F-Stops are exponential. Each full F-Stop increment represents a 100% increase in light transmission from the previous stop.

In the case of M-X vs. M, there really isn't an increase in range due to sensitivity or not much of one. It's about making better use of the information that is already there. With the M sensor, we have 12+ stops of range, but how much is actually usable due to the signal to noise ratio? 9.5 to 10 stops in most cases? With M-X we are gaining 1.5 to 2 stops of usable range through an improved signal to noise ratio, but are not losing any precision because the overall range itself hasn't been altered much in regards to the 12bit luma scale.

We also have to be careful about how we reference "stops" in relation to an imaging system. it's a language that camera professionals like to speak in, yet I don't think many understand it as well as they think they do. Stops are a non-linear, abstract unit of measure. To say we have gained 2 stops of usable latitude can mean different things, depending on just where those two stops are gained. If they are gained on the low end of the luma scale, by reducing our acceptable noise floor, we have not gained a huge percentage of overall dynamic range. If we gain 2 stops on the high end of the luma scale, that could be analogous to increasing the DR by up to a factor of 4.
 
Jeff is certainly correct that the allocation of values from the "bit-range" selected to record an arbitrary number of luminence "stops" aren't directly correlated.

It's really coincidence that each increasing F-stop doubles the amount of light represented, and each additional bit added to a digital "word" doubles the number of uemerical values that word represents.

A 12 bit word (as mentioned) represent 4096 values. So you spread these vales out to describe watever overall range of light you want to represent. You have ~4000 "steps" to transition from the lightest light you can record to the deepest dark. All you need is enough steps to do this without banding (i.e. too large of a jump in between steps).

It so happens that as our eyes get more overwhelmed by light, they are less sensitive to small variations. (So, while you might be able to discern the difference between 10 and 20 photons/sec hitting your retina, you can't discerns the difference between 100,010 and 100,020/sec).

Thus you simply have to be careful on the low end. Provided you have enough steps, you can distribute them linearly and not not have to make the jumps power of 2.

Really, RED doesn't have to do anything, as the top end is fixed. That is the same number of photons/sec causing a M sensor to register value 4096 on the scale is also what will cause M-X to register 4096 (aka "full white"). It just so happens that at the low end, an value below 400 was considered too noisy. Now you can record down to any value below 250 before it's too noisy. You have now increased your usable recorded DR. However those values at the low end have been recorded along.

Really, while refering to the DR a system can capture in terms of bits is a handy shorthand (especially for engineers), I suspect a lot of people would be less confused if it were referred to in decimal or some other notation... just to get the "power of 2" confusion out of there.

-sc
 
Pardon me if my I am misunderstanding, but I can see how people are getting confused by this. Jeff you agree that the sensor captures light linearly and that light is measured exponentially. I'm fairly sure you do need over 13 bits of A/D to capture over 13 stops. If 12 bits allows us from 0-4096 raindrops, and rainstops are a square exponential measurement of raindrops (raindrops = 2^rainstops), and only 11 rainstops worth of rain fall in any cup, the highest value measured will be 2048, right? what happens if 13 rainstops aka 8192raindrops fall in a cup?

Sounds like some of the confusion is about A/D precision vs recorded precision? I am under the impression you need at least as many bits of precision as you want stops of dynamic range when converting photosite voltages to numerical digital values, but then you can encode those linear values however you want to your codec stream. My guess is that RED IS using 15bit A/D for their newer cameras (anyone know what the precision of the A/D on RED ONE is?) as that IS necessary for achieving >13stops of latitude and that is not log encoded but perhaps gamma encoded into 12bit REDCODE RAW. I think this theory that REDCODE is gamma corrected is supported by the fact that redcode vbr bitrate goes up when the cap is on.(1)

(1) http://provideocoalition.com/index.php/mcurtis/story/red_one_geekery_real_world_info_on_redcode/
 
The sensor captures light as linear analog voltage. If the analog range of the sensor between clipping and noise floor is 13 stops you have 13 stops to play with. It is then converted from analog to digital in discrete steps with 12 bit AD conversion providing 4096 discrete value steps between noise floor and clipping over the full 13 stop range. If there is not enough bit depth to represent the full range accurately, then you get artifacts like banding where the steps are too far apart to smoothly resolve fine shifts in value at low contrast. So as the analog DR goes up, you need increased bit depth for smooth gradation of fine changes in analog values. But it is the original analog value range that determines ultimate DR limits.
 
You must log in or register to reply here.
Back
Top