Mysterium-X...

[Tim Whitcomb]

Short answer, no. More ND will give you more highlight protection. Same clipping penalty for turning down the ISO.

When we put a curve on an image, we are deciding where the image will appear dark and where it will appear bright. Just think of changing the ISO as a decision between more shadows or more highlights.

Lower ISO = More of you image signal devoted to shadows. Less of your image signal devoted to highlights. Less noise, easier to clip.

Higher ISO = More of your image signal devoted to highlights. Less of your image signal devoted to shadows. More noise, harder to clip.

The image signal isn't changing, unless you change the actual amount of light hitting the sensor and it will still clip at the exact same place (roughly).

Since M-X has less noise overall, than the decision to use a higher ISO is more likely. In bright situations you will need more ND so that you have that choice.

IBloom

17 pages. 151 posts and I FINALLY understand this thread!

Thanks IAN!

 

[I Bloom]

Yepp, thats what I ment
IR not UV...
The DoP that got this problem said that D21 (or maybe F35) doesn't suffer from this problem, cus it has a "built in Hot Morror" on the sensor to block the IR.
Any ideas of doing this on the M-X?

/carl

One last diatribe and then I'll retire for a few months:

The Red One has a built in IR filter, you can see it about about 8mm infront of the sensor. This filter is specifically tuned to take out the right amount of IR so that your image is not polluted under normal light sources (I think sunlight is probably the the most balanced.)

Problems only arise when one places an absorptive ND filter infront of the lense. The absorbtive filter blocks light in the optical spectrum but allows infra red light to pass through, so while you are properly exposing the optical spectrum you are overexposing the sensor with IR light. This overexposure overwhelms the cameras internal IR filter. The result can be milky blacks, weird color effect on some materials and in extreme cases additional Red channel exposure on certain materials that can for example make green leaves turn brown.

The solution is to use different kinds of filters to block out the additional exposure of IR light. There are two methods that I know of... I use both.

Use a Mirror NDs:
They aren't all created equal but the idea behind the mirror is that the metalic coating reflects as much IR light as it does optical light, thus eliminating the IR overexposure problem. I highly recommend Panchro Mirror NDs. They are telescope grade and make beautiful images. Unfortunately actors tend to look into them

Use an IR Cut filter in front of your absorptive NDs:
and thus compensate for the additional IR light that would be reaching your sensor. Again there are a lot of different filters on the market, best to test them all. I have two Schneider TrueCuts for my cameras.

I can't speak for the D21, but I know that all digital cameras have this issue simply because silicon is as sensitive to IR as it is to optical light. Camera designers must walk a tight balance between enough filtration and avoiding unwanted reflective elements behind the lense, and other side effects of the various filtration materials. If the designer can find a material that somehow creates a better balance then great, but perhaps it's better not to expect that the camera needs to change to work with tools designed for film, instead just adapt your tools to the medium of silicon.

IBloom

 

[Shawn Nelson]

Short answer, no. More ND will give you more highlight protection. Same clipping penalty for turning down the ISO.

When we put a curve on an image, we are deciding where the image will appear dark and where it will appear bright. Just think of changing the ISO as a decision between more shadows or more highlights.

Lower ISO = More of you image signal devoted to shadows. Less of your image signal devoted to highlights. Less noise, easier to clip.

Higher ISO = More of your image signal devoted to highlights. Less of your image signal devoted to shadows. More noise, harder to clip.

The image signal isn't changing, unless you change the actual amount of light hitting the sensor and it will still clip at the exact same place (roughly).

Since M-X has less noise overall, than the decision to use a higher ISO is more likely. In bright situations you will need more ND so that you have that choice.

IBloom

This is a great post, and yet I feel stupid reading it because this is what I dont understand:

On M-X, if I turn ISO from 800 to 250 without changing ANYTHING else, will that preserve a bunch of highlights that would have otherwise been lost or is it merely changing the perspective? On Red this is an easy answer, you're simply looking at it differently, you arent preserving anything, which is what sucks, you arent actually changing iso as you are on a DSLR or with film. Hopefully changing ISO on M-X actually changes the sensitivity of the sensor and not merely metadata, I need more actual highlight protection instead of just a different perspective on the RAW data.

 

[Blair S. Paulsen]

The way I understand it is:

The way I understand it is:

I would be surprised if M-X is any different in this area than Mysterium, other than lower noise. Unless RED decides to add analog gain circuitry, which would be a significant change in philosophy, all ISO discussion is simply about the shooters mental approach to setting exposure levels -

the characteristics of the sensor remain the same no matter what ISO number you use during shooting.

If your post operations never alter any of the on camera meta-data settings then you are voluntarily treating that ISO setting as a "baked in" value but it is not.

The part that can be confusing: based on the nominal ISO set during shooting the operator may look at the monitor output and choose to change the aperture or add filtration, it is that act, not the ISO number, that actually changes what is captured. When someone recommends upping the ISO on a bright exterior it is to prompt the operator to lower the actual exposure, it does not make the sensor any more or less sensitive and it does not alter the image data set created, just the meta-data.

 

[Jagger Christian]

Just heard the announcement of the CANON 1D Mark IV and saw Vincent LaForet's Short Film with the camera. This is getting interesting in terms of ISO. I wonder how the new Scarlet & Epic are going to look in comparison.

 

[Alex Mack]

Just heard the announcement of the CANON 1D Mark IV and saw Vincent LaForet's Short Film with the camera. This is getting interesting in terms of ISO. I wonder how the new Scarlet & Epic are going to look in comparison.

This was my thinking as well.
I hope MX can come close...

That short was exactly my spirit of movie making.
Dark, subtle natural lighting.

 

[Tim Whitcomb]

Just heard the announcement of the CANON 1D Mark IV and saw Vincent LaForet's Short Film with the camera. This is getting interesting in terms of ISO. I wonder how the new Scarlet & Epic are going to look in comparison.

HUH?? http://www.vincentlaforet.com/index_reverie.html
?? perhaps its web compression??, but this looks like cheap video to me... artifacts, aliasing... motion judder. Noise galore. The lighting is nice, but this is not a great representation of how filmic-motion should look.

I trust the M-X will blow this out of the water.

Remember. DSMC = Still and Motion = SAME high quality.

 

[Alex Mack]

HUH?? http://www.vincentlaforet.com/index_reverie.html
?? perhaps its web compression??, but this looks like cheap video to me... artifacts, aliasing... motion judder. Noise galore. The lighting is nice, but this is not a great representation of how filmic-motion should look.

I trust the M-X will blow this out of the water.

Remember. DSMC = Still and Motion = SAME high quality.

He did say in terms of ISO.
The 1D suggests a future of film making - super high iso - which could come from any maker at any time (most likely red, and hopefully in MX).

Edit: I see now, you were thinking 5d reverie.
He meant this, since last night:
http://vincentlaforet.smugmug.com/L...IV-Video/10024122_sqhwE#686345820_EeDCa-X2-LB
Filmic motion (180 degree shutter) at iso 6400. No lighting. Don't you just love it?

 

[Carsten Kurz]

5DmkII,7D,1D Mark IV in video mode suck in literally all aspects against RED, but not in sensor size and lowlight capability. Great cams, but still enough sucking to not compare them with RED.

The video on that link is too heavily compressed to judge. There's a lot of impressing Canon stuff at Vimeo.

Back on topic: With regards to 'sensivity' - what is canon doing differently when changing ISO? As Mysterium, it is CMOS. Do they use analog gain stages or A/D adjustments? They have 'real' ISO setting in RAW stills and video. In video mode, binning will reduce sensor noise somewhat additionally, of course (which is the only positive aspect of the Canon binning). The DSMCs full frame 1080p modes will benefit from that effect as well.

I guess that pro/con analog gain for Mysterium has been discussed on this forum before. Any hints?


- Carsten

 

[I Bloom]

This is a great post, and yet I feel stupid reading it because this is what I dont understand:

On M-X, if I turn ISO from 800 to 250 without changing ANYTHING else, will that preserve a bunch of highlights that would have otherwise been lost or is it merely changing the perspective? On Red this is an easy answer, you're simply looking at it differently, you arent preserving anything, which is what sucks, you arent actually changing iso as you are on a DSLR or with film. Hopefully changing ISO on M-X actually changes the sensitivity of the sensor and not merely metadata, I need more actual highlight protection instead of just a different perspective on the RAW data.

All you have to do to get that highlight detail is stop down. A sensor is more sensitive because it has better information about less light. More sensitive sensors don't really collect more light, they just do a better job of measuring the light that is there.

Likewise I don't think you are actually changing the sensitivity of the camera on a DSLR. You just think you are because the camera is delivering an image with an applied curve (either in the camera or in the software that imports the raw images.) I think you'll find that the exact same conditions apply to every CMOS camera if you can get access to the actual raw data.

I think it's unfortunate that we've been trained over the years by video cameras that apply a look to the image to believe things are the other way around. I'm very happy about the raw image that the Red camera produces and the fact that the rawness survives compression.

I think it's unfortunate that Canon seems to be doing just the opposite, baking a look into an image from a DSLR sensor and then dumping it into a video codec. I see that as basically a crowd pleaser move that I think will translate to a set back for the artistry of cinematography when we are all forced to shoot commercials on their upcoming HD Video-DSLR hybrid... part of the reason that I work on raw workflow tools.

Here's to hoping raw rules the day.

IBloom

 

[Jason Diamond]

Extremely well put and the most concise explanation of the IR issue.
Thanks Ian....!!!

 

[Sanjin Jukic]

All you have to do to get that highlight detail is stop down. A sensor is more sensitive because it has better information about less light. More sensitive sensors don't really collect more light, they just do a better job of measuring the light that is there.

Likewise I don't think you are actually changing the sensitivity of the camera on a DSLR. You just think you are because the camera is delivering an image with an applied curve (either in the camera or in the software that imports the raw images.) I think you'll find that the exact same conditions apply to every CMOS camera if you can get access to the actual raw data.

I think it's unfortunate that we've been trained over the years by video cameras that apply a look to the image to believe things are the other way around. I'm very happy about the raw image that the Red camera produces and the fact that the rawness survives compression.

I think it's unfortunate that Canon seems to be doing just the opposite, baking a look into an image from a DSLR sensor and then dumping it into a video codec. I see that as basically a crowd pleaser move that I think will translate to a set back for the artistry of cinematography when we are all forced to shoot commercials on their upcoming HD Video-DSLR hybrid... part of the reason that I work on raw workflow tools.

Here's to hoping raw rules the day.

IBloom

BTW, do you remember Profile: SC 4.1 250ISO???

 

[Daniel Browning]

On M-X, if I turn ISO from 800 to 250 without changing ANYTHING else, will that preserve a bunch of highlights that would have otherwise been lost or is it merely changing the perspective? On Red this is an easy answer, you're simply looking at it differently, you arent preserving anything, which is what sucks, you arent actually changing iso as you are on a DSLR or with film. Hopefully changing ISO on M-X actually changes the sensitivity of the sensor and not merely metadata, I need more actual highlight protection instead of just a different perspective on the RAW data.

Let me see if I can help explain the situation.

Changing the ISO setting on a DSLR does not change the sensitivity of the sensor. Such a feature would be nice, but does not actually exist in any of the existing cameras yet.

So given that you have a sensor that collects a certain amount of light, no matter what, there are only three things you can do to change the brightness in a raw camera:

• Metadata gain
• Useful analog gain
• Worthless analog gain
• Idiot gain

Metadata gain is the only one that the RED ONE uses: they do not have user-controllable analog gain or idiot gain. In this type of gain, the data from the sensor is left as-is until post.

Analog gain increases brightness. But if that was all it ever did, then no one would ever use it. It would be worthless analog gain. But there are certain circumstances where it does *more* than just increase brightness. It actually *reduces* noise. That's when it becomes useful analog gain. On a lot of DSLR, especially Canon, the late-stage electronics (ISO amplifier, ADC, etc.) have so much noise that if you don't use analog gain, they will contribute a lot of extra noise to the signal. As you increase the analog gain and feed those electronics with a stronger and stronger signal, their contribution to the total noise gets smaller and smaller.

For example, here is the raw data from my Canon 5D2 at ISO 100:

And here is the same exact exposure but at ISO 1600:

The signal to noise ratio (SNR) is 1 and 1/3 stops better. But there is a cost! ISO 1600 clips *four stops* of highlights compared to ISO 100 for the same exposure.

My guess is that the RED camera designers are smart, and they tested the option of offering user-controllable analog gain. If so, they must have found that there was no improvement in noise. This may have been because their late-stage electronics (amplifier, ADC, etc.) are much less noisy than Canon's. Or it could also be there is already enough noise to drown out the noise added by the ADC. Either way, the result is that there would be no benefit to having analog gain on RED ONE.

The practical reality, for most CMOS DSLRs so far, is that gain is needed to reduce noise. But the optimal situation will be to have the same noise but with no analog gain at all. Maybe that will be possible in the future. (The D3X is a step in the right direction.)

Idiot gain is digital gain applied to the raw file in-camera. It doesn't have the benefit of analog gain, which sometimes, under some circumstances, can reduce noise. But it does have the drawbacks of analog gain: it clips highlights. It also has other drawbacks, including increased file size and posterization.

Of course you might think that no camera designer in their right mind would ever use idiot gain, and instead rely on metadata gain or analog gain. The unfortunate reality is that most DSLRs have idiot gain. For example, on my 5D2:

ISO 50: metadata gain
ISO 100: no gain
ISO 125: idiot gain
ISO 160: idiot gain
ISO 200: very useful analog gain
ISO 250: idiot gain
ISO 320: idiot gain
ISO 400: medium useful analog gain
[...]
ISO 800: kinda useful analog gain
ISO 1600: slightly useful analog gain
ISO 3200: useless analog gain
ISO 6400: idiot gain
ISO 12800: idiot gain
ISO 25600: idiot gain

All of the settings that use idiot gain should of course use metadata gain instead, but they don't. ISO 25,600 actually clips four stops of highlights compared to ISO 1600 with the same exposure, yet has the exact same noise level (SNR)! The only metadata gain on my camera is ISO 50. HTP is a kind of metadata gain, but it only has 1 stop.

As to your first question, I don't know if RED will start to offer user-controllable analog gain in the new cameras or not, but if they do it will only be because it reduces noise, not because it icreases highlight headroom (it can't).

What I'd rather see is for DSLR manufacturers to learn from RED (or even MFDB and digicams) and start offering metadata ISO. At least as a replacement for their existing idiot gain, but hopefully also as an alternative to all the existing analog gain settings (especially 400 through 3200, which have varying levels of utility).

 

[david farland]

Daniel,
All amplifiers unless they have filtration built in will amplify both signal and noise. Filtering will change the transfer/bandpass characteristics of the input signal. Amplifiers usually have an input noise threshold and peak (or shoulder) point where you'd like to keep your input signal between. Without the D2 schematic I couldn't comment on how it filters across iso settings.
D

 

[Daniel Browning]

Without the [5D2] schematic I couldn't comment on how it filters across iso settings.

There is no filtering involved at all.

Daniel,
All amplifiers unless they have filtration built in will amplify both signal and noise.

That is correct when the only source of noise is upstream from the amplifier, such as when looking at the noise in midtones at ISO 100 (which is only photon shot noise; nothing else). However, that is very wrong when the amplifier itself and other electronics also contribute to the noise (such as deep shadows at ISO 100 or midtones at high ISO).

Say you have a junky microphone, a good amplifier, and a terrible ADC. If you feed the microphone with a very loud signal (say, -3db), the self-noise added by the microphone and ADC will be small enough that you'll still get a good SNR. Now feed the mic with a weaker signal (say, -36db) but continue to leave the amplifier gain at the same setting. The noises of the microphone and ADC will be soo high that they ruin the SNR. Replacing the microphone with one that has lower self-noise wouldn't help, because you still have the high noise of the ADC. However, if you turn up the gain on the amplifier so that the -36db signal feeds into the ADC as a -3 db signal, now the ADC noise goes back to the level it was before.

That is the same reason why turning up the gain on DSLRs sometimes (but not always) improves the SNR.

 

[david farland]

Observing the final output isn't enough as unwanted responses will be filtered out, so I figure you must have the schematics with transfer characteristics of chip/stage (cds,pga, clamp circuits etc etc ) to make the statement. Also can't follow you mike/adc/amp story but anyway simply put....amplification is used to keep the signal above noise floor, where ever it originates from, distortion is something else and is treated differently.
Dave,

 

[Steven Caesare]

(Warning: long post ahead)

It seems that some source of confusion here is based on the idea (fear?) that if the sensor ISO increases significantly, that everybody will either have to stop lenses down significantly (and thus lose DoF), or stack an insane amount of ND filters in front of the lens.

I don't think this is the case.

It occurs to me that this is similar to the issue of contrast ratio with projectors. CR is the difference between darkest material and brightest material that can be displayed.

The expectation is is that if the contrast ratio for a projector goes up, you can display even brighter material without clipping than you could before.

Nope.

The output of a projector is limited on the upper end by the output of the bulb. The brightest white is simply the bulb at full on. The increase in contrast ratio comes from lowering the lumen output at the bottom end of the darkest material that can be displayed. Ie- "blacker blacks".

I suspect the situation with M-X is similar. At the top end, the sensor will clip at the exact same place. Ie- if it took one million photons to stimulate a M sensor pixel to reach full charge ("full white") , then it will also take one million photons to fully stimulate a M-X pixel. Your top end clip point won't change.

However... on the bottom end, it may have taken 500 photons to stimulate a M pixel with enough charge to acceptably rise above the noise level. On M-X it now only takes 200 photons to stimulate a pixel above the noise level.

Hence the overall range has expanded since the bottom end has moved downward. Therefore, the point at which your mid-gray point lives has also moved downward, dictating an increased ISO rating. (And also illustrating why Jim & Co. have suggested that a single number to rate sensitvity is an approximation that only holds in a specific instance).

The net effect is that you won't clip any more than before at a given intensity level. You will, however, be able to shoot in overall darker conditions before the noise becomes objectionable. Or push in post even farther.

Or I could be completely wrong.

-sc

 

[Daniel Browning]

Observing the final output isn't enough as unwanted responses will be filtered out,

You are mistaken. Again, there is no filtration on 5D2. The Fourier transform of the raw data shows noise power increasing linearly with spatial frequency at all ISO settings. If you look at JPEG, sure, there is plenty of noise filtration especially at high ISO. But the raw data has no median filter, at least in all Canon DSLRs up to the 5D2. There are a few DSLRs, like the Pentax K20D, that do apply a noise reduction filter at certain ISO settings (only ISO 1600 in that case), and it shows up clearly in a FFT. If you have any sort of proof otherwise, I'd love to see it.

...so I figure you must have the schematics with transfer characteristics of chip/stage (cds,pga, clamp circuits etc etc ) to make the statement.

I can do something far better: measure the *actual* performance of the camera itself. It's quite simple in fact. Select several subjects of even brightness (e.g. patches on color checker chart), measure the SNR at all ISO settings, multiply by 1/ISO, and plot std dev vs signal. The intercept is the read noise squared and the slope is the inverse gain. When you do this for the 5D2, you get the following read noise:

ISO 100: 23.5 e- read noise
ISO 200: 11.9 e-
ISO 400: 6.4 e-
ISO 800: 3.7 e-
ISO 1600: 2.5 e-
ISO 3200+: 2.5 e-

That is why increasing the analog gain results in less noise and higher SNR for a fixed exposure, as the images I posted above demonstrate.

 

[I Bloom]

That is why increasing the analog gain results in less noise and higher SNR for a fixed exposure, as the images I posted above demonstrate.

I'm not sure I understand where you are going with this. If the clipping point is a fixed property of the chip then why isn't there one analog gain setting that is ideal for all shooting situations?

IBloom

 

[Emmanuel Cambier]

(Warning: long post ahead)

It seems that some source of confusion here is based on the idea (fear?) that if the sensor ISO increases significantly, that everybody will either have to stop lenses down significantly (and thus lose DoF), or stack an insane amount of ND filters in front of the lens.

I don't think this is the case.

It occurs to me that this is similar to the issue of contrast ratio with projectors. CR is the difference between darkest material and brightest material that can be displayed.

The expectation is is that if the contrast ratio for a projector goes up, you can display even brighter material without clipping than you could before.

Nope.

The output of a projector is limited on the upper end by the output of the bulb. the brightest white is simply the bulb at full on. The increase in contrast ratio comes from lowering the lumen output at the bottom end of the darkest material that can be displayed. Ie- "blacker blacks".

I suspect the situation with M-X is similar. At the top end, ie sensor will clip at the exact same place. Ie- if it took one million photons to stimulate a M sensor pixel to reach full charge ("full white") , then it will also take one million phtons to fully stimulate a M-X pixel. Your top end clip point won't change.

However... on the bottom end, it may have taken 500 photons to stimulate a M pixel with enough charge to acceptably rise above the noise level. On M-X it now only takes 200 photons to stimulate a pixel above the noise level.

Hence the overall range has expanded since the bottom end has moved downward. Therefore, the point at which your mid-gray point has also moved downward,dictating an increased ISO rating. (And also illustrating why Jim & Co. have suggested that a sinngle number to rate sensitvity is an approximation that only holds in a specific instance).

The net effect is that you won't clip any more than before at a given intensity level. You will, however, be able to shoot in overall darker condtions before the noise becomes objectionable. Or push in post even farther.

Or I could be completely wrong.

-sc

Very well put…

Thanks a lot.

Emmanuel