Welcome to our community

Be a part of something great, join today!

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 :)

Problem... solution.

Because the two greens don't give higher green sensitivity, but more green (close to luma) resolution.

Graeme
 
In the blue part of the spectrum, that graph certainly shows more sensitivity, but I think the relative difference in sensitivity between blue and red/green is actually getting worse, not better, with the new black silicon. How do you read it? I see about normal silicon on their graph being about 5:1 for r/g : b, and their black silicon around 12:1.

Graeme
 
Its definitely more biased. I figured that could be controlled (damped) electronically to achieve a flatter response curve.
If I use the white paper, blue maps from about (ruler on screen!)
0.1 to 1.1 at 425-450nm so about 11x ish.
.3 to 10.1 at around 500nm around 34x ish.
average is about 22.5 but a blend would give well over 25X (area under curve).
 
For underwater work, you wouldn't need to damp the Red much as there is so little. In fact you would want the bias I think. It would be pretty flat and just need some minor colour correcting (depth dependent etc.). I figure white balance would "pivot around green" as usual.
 
I have successfully shot Red under Tungsten light at 400ASA. The key is not to expect useful information (aka noisy shadows) where there is none. The danger of exposing from a small LCD monitor (EVF/Onboard) is that it masks noise.
I come from a film background and have always lived by the zone system. Just about any imaging system goes to black at around 4 stops under and to white at about 3.5 over when the distribution format is accounted for.
RedRaw as well as film will record information beyond that in both directions on the Digital Negative/Negative and it is up to the DP to determine what portion of the originally recorded image is acceptable to be used in the final distribution image.
Oversampling is the key to a strong final image in any imaging system. One must expose with the final image in mind, not the oversampled intermediate. You can use the oversampled intermediate information in a pinch when an exposure mistake has been made, but it will result in inconsistent and heavily treated, less honest and direct imagery.

This subject also illuminates why some early dynamic range tests have been so vastly different. I bet some have been shot under Tungsten light.

I use an 80D for greenscreen shot under Tungsten but haven't seen the need to do so for regular Tungsten work. I am quite fond of Tungsten light and am not quite ready to abandon it for good.

Here's some samples of Red build 15 @ 400ASA under Tungsten

tango1.jpg

tango2.jpg

tango3.jpg

tango5.jpg
 
It seems that the easy way to understand this, for the proverbial lay man, is that if you want/have to use tungsten on a subject, try to use daylight balanced lighting for all your fill light needs. This will provide blues on the less exposed areas, the ones that need it most, and also increase the contrast between cool and warm tones from the subject to the background, which should be a good thing. This is what I've done with the F900 and other cameras, from the Ex1 to the minuscule Canon HG20, and works well, though then again, I've never been afraid to crush those blacks to eliminate noise if needed.

Underwater shooters don't to worry about this issue. There is nothing BUT blue light underwater, as close to the surface as 3 meters/10 feet, so you should get very clean images there, provided you light up the subjects you're shooting to prevent an overall bluish cast to the image. I prefer to use warm lights (tungsten like) to light up subjects since this makes the contrast between them and the surrounding "blueness" all the more dramatic, although the newer HID lights (daylight balanced) offer a more natural gradation of colors.
 
Roberto Lequeux said:
In post?
Click to expand...

No, not in post! You can't really "add" blue light in post, all you can do is try and raise the blue channel which will only increase the noise. You need to try and add as much blue light as possible (while filming) to those areas that will need it most, meaning, dark areas. The better lit areas can get away with a lot more, but unless you have enough blue on those darker spots, you'll end up with noise or the need to crush the images substantially.
 
Brent J. Craig said:
Would it make more sense for a cinema camera to be natively balanced for tungsten?
Click to expand...

As David pointed out, this can't be done with current sensors. Even if they were somehow able to get the quantum efficiency of blue up to the same level as green, it would *still* not be balanced for tungsten, because tungsten light itself has very little blue in it.

A sensor cannot see something that isn't there. And the blue light just isn't there. It would be necessary to put a 1-stop ND on the green/red filters to get them down low enough so that the sensor would be "native" tungsten. Obviously, that would serve no purpose; we can already use an 80C filter to do that.

As technology advances, there will be improvements to read noise, which will benefit tungsten as well as all other shooting circumstances. As David pointed out, once the read noise is good enough, the problem is effectively solved (no matter the "native" balance of the sensor).

James Derek Holloway said:
High hope for Black Silicon in the future...almost 100x more sensitive! 6+ stops has advantages and complications.
Click to expand...

Their use of the word "sensitive" in marketing materials is probably different than what you and I think of as "sensitive".

No technology, not even Black Silicon, can see light that isn't there. If you put the lens cap on, no amount of sensitivity will let it see light. Similarly, once you detect 100% of the light that falls on the sensor, that's it. Do not pass go. Do not collect 200 photons.

Current technology, e.g. cheap digicams, can already detect 70% of the light that hits the sensor. The maximum theoretical improvement is another 30% (though it's not practically possible, it's at least theoretically possible).

As camera development has gotten closer and closer to the theoretical limit, the improvement in sensitivity has slowed greatly. Looking at Canon DSLR, for example, QE has improved 5% in the last 1 year. 30% in the last 3 years, and 115% in the last 5 years. [10D: 0.042 photo-electrons per 12-bit ADU per square micron. 50D: 0.100 e-/adu/um^2.]

I think this signifies that Canon is hitting the QE limits. Black silicon can't get around that. However, it's possible that it will improve read noise, and therefore "signal to noise ratio", or S/N.

Canon biggest improvement in the last 5 years was high-ISO read noise. Combined with the QE improvement it allowed the high ISO shadows to improve S/N by 3.2 stops from 10D to 50D. (using S/N at 8 stops below clipping for 12 micron CoC).

Base ISO read noise and full well capacity improvements over the same time period improved base ISO dynamic range by just 0.77 stops (not counting pattern noise).

Luckily, there is lots of room for improvement in read noise and no theoretical boundaries. Of course, there are tons of practical limitations, which is why sensor designers have such a hard job. My point is that if Black Silicon will have any significant low light benefit, it will be due to the reduction of read noise, not an increase in sensitivity; because, again, the current technology is already at the theoretical limit.

The difference between true sensitivity ("quantum efficiency") and read noise is important. For example, if RED improves read noise in EPIC S35 by 2 stops (and no change to QE), we'll be able to shoot ISO 1280 and get the same level of shadow noise as we get now at ISO 320. However, the midtones and highlights will actually have *more* noise. But given how clean they are now, I wouldn't mind more noise in those exposure zones, and I think other photographers wont either.

I hope that helps illustrate the difference between QE and read noise. The "suggested rating" (or "sensitivity") of 320 ISO for the RED ONE is based on a mix of these two factors.
 
Jeff Kilgroe said:
Have you talked to RED, given them a clip to look at, etc..?
Click to expand...

I have not directly talked to them direcly. Have posted samples and posted comments in multiple threads. Both my cams do it -- and I'm not the only one. Here's a different thread -

-http://www.reduser.net/forum/archive/index.php/t-15913.html

I assumed that all Reds do it, but hoped it would be fixed in a firmware update.

Do you not see the vertical lines in underexposed media from your cam? You would know it if it's there, especially if shot tungsten.

It is not a worry for shooting under controlled conditions -- only low light available light scenarios.
 
Doc Holliday said:
Do you not see the vertical lines in underexposed media from your cam? You would know it if it's there, especially if shot tungsten.
Click to expand...
I notice this in dark areas even on well-exposed footage. It's usually an easy case to just crush the blacks and rid yourself of the problem, but the grading doesn't always call for that. Noise can't always be avoided, but the fixed patterns are frustrating.
 
Daniel,

No question we abuse the word sensitivity to-no-end. And read noise is definitely part of the picture. Thin layers will help this as well at low light. Much less dispersion etc.
 
Daniel Browning said:
As David pointed out, this can't be done with current sensors. Even if they were somehow able to get the quantum efficiency of blue up to the same level as green, it would *still* not be balanced for tungsten, because tungsten light itself has very little blue in it.

A sensor cannot see something that isn't there. And the blue light just isn't there. It would be necessary to put a 1-stop ND on the green/red filters to get them down low enough so that the sensor would be "native" tungsten. Obviously, that would serve no purpose; we can already use an 80C filter to do that.

As technology advances, there will be improvements to read noise, which will benefit tungsten as well as all other shooting circumstances. As David pointed out, once the read noise is good enough, the problem is effectively solved (no matter the "native" balance of the sensor).



Their use of the word "sensitive" in marketing materials is probably different than what you and I think of as "sensitive".

No technology, not even Black Silicon, can see light that isn't there. If you put the lens cap on, no amount of sensitivity will let it see light. Similarly, once you detect 100% of the light that falls on the sensor, that's it. Do not pass go. Do not collect 200 photons.

Current technology, e.g. cheap digicams, can already detect 70% of the light that hits the sensor. The maximum theoretical improvement is another 30% (though it's not practically possible, it's at least theoretically possible).

As camera development has gotten closer and closer to the theoretical limit, the improvement in sensitivity has slowed greatly. Looking at Canon DSLR, for example, QE has improved 5% in the last 1 year. 30% in the last 3 years, and 115% in the last 5 years. [10D: 0.042 photo-electrons per 12-bit ADU per square micron. 50D: 0.100 e-/adu/um^2.]

I think this signifies that Canon is hitting the QE limits. Black silicon can't get around that. However, it's possible that it will improve read noise, and therefore "signal to noise ratio", or S/N.

Canon biggest improvement in the last 5 years was high-ISO read noise. Combined with the QE improvement it allowed the high ISO shadows to improve S/N by 3.2 stops from 10D to 50D. (using S/N at 8 stops below clipping for 12 micron CoC).

Base ISO read noise and full well capacity improvements over the same time period improved base ISO dynamic range by just 0.77 stops (not counting pattern noise).

Luckily, there is lots of room for improvement in read noise and no theoretical boundaries. Of course, there are tons of practical limitations, which is why sensor designers have such a hard job. My point is that if Black Silicon will have any significant low light benefit, it will be due to the reduction of read noise, not an increase in sensitivity; because, again, the current technology is already at the theoretical limit.

The difference between true sensitivity ("quantum efficiency") and read noise is important. For example, if RED improves read noise in EPIC S35 by 2 stops (and no change to QE), we'll be able to shoot ISO 1280 and get the same level of shadow noise as we get now at ISO 320. However, the midtones and highlights will actually have *more* noise. But given how clean they are now, I wouldn't mind more noise in those exposure zones, and I think other photographers wont either.

I hope that helps illustrate the difference between QE and read noise. The "suggested rating" (or "sensitivity") of 320 ISO for the RED ONE is based on a mix of these two factors.
Click to expand...


Excellent post. Thanks.
 
I've also noticed that your processing method can greatly change the effect. For 4k footage, 1/4 res and Full res look the best, while half res seems to make it more apparent.


Häakon said:
I notice this in dark areas even on well-exposed footage. It's usually an easy case to just crush the blacks and rid yourself of the problem, but the grading doesn't always call for that. Noise can't always be avoided, but the fixed patterns are frustrating.
Click to expand...
 
Talking about blue channel sensitivity, as a researcher in vision with an interest in cinematography, the human fovea has a similar relative insensitivity to blue spectrum light, in this case due to sampling bias:
attachment.php
Roorda et al., 2001 doi:10.1016/S0042-6989(01)00043-8

This shows relative red, green and blue cone receptors at about 1 visual degreee away from the centre of vision (a very high resolution sampling area called the fovea) for 3 humans and one monkey. Actually there are basically no blue cones at the fovea, we a tritanopic (blue colour blind), with the very highest resolution part of our vision! Blue receptors are on average only 5% of the total number of cones, and yet we do not see blue as so significantly less visible. It appears the brain uses analogue gain as a solution, and there is a separate stream called the koniocellular channel that deals with blue light information into the visual system. We do know that reaction times are slower for blue than red/green, and absolute senstivity at levels that are at the limits of our sensitivity are clearly different.

Also interesting, blue receptors (S or short-wave cones) clearly predate the evolution of mammals and the gene for the blue pigment is on a different chromosome (7) to red and green pigments (X chromosome).

I'm pretty sure there are additional perceptual mechanism apart from the gain to compensate for the weak blue channel that feeds our visual system, and it wouldn't surprise me if we have differential noise sensitivity, effectively minimising the blue channel noise our visual systems must face. As an additional aside, it is also interesting to note that noise often enhances visual processing, and can substantially improve the efficiency with which our brains can encode visual information. So, some noise may therefore be advantageous for our perception of moving images!
 
Daniel Browning said:
No technology, not even Black Silicon, can see light that isn't there. If you put the lens cap on, no amount of sensitivity will let it see light. Similarly, once you detect 100% of the light that falls on the sensor, that's it. Do not pass go. Do not collect 200 photons.
Click to expand...

One mechanism they describe in their paper is photo conductive gain. While current production sensors may well be approaching 90% efficiency in some wavelengths, they are only producing one electron pair. Very difficult to detect. In a photo conductive gain system, you may well get your 200 electron pairs per photon. This would definitely increase the chance of photon detection and reduce read noise from various sources (dark current? etc) Actually this reminds me of lasers. You use smaller lasers to trigger doped glass to produce a massive laser bursts; only you're producing more photons not electrons. This would be an interesting system though. 5K Microlaser panels....:sifone:

drl.ee.ucla.edu/images/PublicationsImages/Photoconductive%20Gain%20SiGeSi.pdf

Another publication
 
You must log in or register to reply here.
Back
Top