What happens as we turn the ISO down?

[Stephen Pruitt]

Hi all. . .

I'm curious as to how and why image quality degrades as we turn the ISO down from 320 to, say, 160 or even 80. This would seem to serve the same purpose as an ND filter (I'm thinking for a mattebox-less steadycam set-up just now) and buy us less noise in the process.

Stephen

 

[Harry Clark]

Stephen,
Forgive me if I'm going over well-trodden ground here.
The most important thing to remember about Red is that the sensor's dynamic range is greatest at 320 ISO. And changing the setting on the camera in no way affects the amount of light hitting the sensor and the RAW file; only the appearance on the monitor is changed.
Rating above 320 runs the risk of severe underexposure; anything in the shadow area will be difficult to pull up without increasing noise. Rating under 320 runs the risk of severe overexposure; anything near the white point will likely be clipped, with no information left to pull down.
I'm GREATLY oversimplifying here. What is the importance of this rule?
If your scene does not have any super bright or specular areas, you can safely underrate. If your scene is well lit with some serious highlights, you can feel free to overrate. Unfortunately, the scenarios are polar opposites to the action required. You'd likely NEED to overrate (pushing or higher ASA film) in a dark, lit-by-practicals night interior. And you would likely wish for a lower ISO (pulling of low ASA film) in the brightest, harshest day exteriors (as I suspect is the case in your application)
Perversely, all our training in the world of photochemical film runs counter to the laws of silicon based imagers.
And again, I'm greatly oversimplifying. There are shades of grey (pardon the pun) and you need to assess each scene by itself to achieve optimum exposure.
In your situation, I'd tape the ND filters directly on the lens... and keep in mind, modern clip-on matteboxes by Arri and Chrosziel are pretty low-profile and lightweight, perfect for Steadicam in even the most cramped quarters. But taping filters on is a well-established practice of hardcore ACs and should not be looked down on in any way.
Cheers,
Harry

 

[Douglas Underdahl]

GREAT post, Harry. Thank you.

 

[tj williams]

Hi Harry

yet when we set our HD camera for -3Db reducing the sensitivity we get less noise and higer \gain ie 9db gives more noise???? also when we need to see the monitor most in low light conditions we have no way to brighten the caamera monitor output to steadicam wo ruining dynamic range.

good way to attach filters use the stringy sticky stuff that cmes on double sided tape along the metal portion of the filter, will always hold.

TJ

 

[Paul Leeming]

yet when we set our HD camera for -3Db reducing the sensitivity we get less noise and higer \gain ie 9db gives more noise????

TJ, that's because you are applying a voltage gain to the ADC itself, meaning that it will become more noisy because the pixels are in effect "overclocked". Red eschews this method in favour of recording RAW and applying software gain in post production, which they determined to be the better option for them.

HTH

Paul

 

[Matthew Verkler]

Hi Harry

yet when we set our HD camera for -3Db reducing the sensitivity we get less noise and higer \gain ie 9db gives more noise???? also when we need to see the monitor most in low light conditions we have no way to brighten the caamera monitor output to steadicam wo ruining dynamic range.

good way to attach filters use the stringy sticky stuff that cmes on double sided tape along the metal portion of the filter, will always hold.

TJ

And, if you need to see the monitor better in low light shooting Red, put the ISO at whatever you need, 1000, 2000, whatever, knowing that it's only metadata and it's not ruining your footage. Also know that what you see on the monitor is not really representative of what you are shooting; trust your meter just like with film.

Matthew Verkler

 

[Joel Kaye]

This would seem to serve the same purpose as an ND filter

An ND filter actually alters the amount of light hitting the sensor. Changing the ISO on RED only changes the metadata. That impacts the viewing system and meters but does not alter what's actually recorded.

Having said that - experimenting has shown that in low dynamic range environments (set interiors) you can rate the camera around 200 and probably produce a better image. Outdoors rating the camera at 200 would likely result in clipping highlights too soon.

Easy enough to test this kind of idea. Go out and shoot with and without an ND at the same F-Stop. Adjust the ISO down when you don't have the ND on and push both images around in RedCine. One will look better than the other.

 

[Kyle Presley]

And, if you need to see the monitor better in low light shooting Red, put the ISO at whatever you need, 1000, 2000, whatever, knowing that it's only metadata and it's not ruining your footage. Also know that what you see on the monitor is not really representative of what you are shooting; trust your meter just like with film.

Matthew Verkler

Yeah, you must expose for the final ISO you assign in RED Alert or RedCine, but view it anyway you want.

 

[martinnoweck]

Hi Harry

yet when we set our HD camera for -3Db reducing the sensitivity we get less noise and higer \gain ie 9db gives more noise???? also when we need to see the monitor most in low light conditions we have no way to brighten the caamera monitor output to steadicam wo ruining dynamic range.

good way to attach filters use the stringy sticky stuff that cmes on double sided tape along the metal portion of the filter, will always hold.

TJ

imo setting gain to -3dB is not the best practice because the light hits the sensor with the same intensity - so using a nd filter would be better.

best wishes,
martin

 

[Sam Roberts]

Stephen, I'd tape the ND filters directly on the lens... and keep in mind, modern clip-on matteboxes by Arri and Chrosziel are pretty low-profile and lightweight, perfect for Steadicam in even the most cramped quarters. But taping filters on is a well-established practice of hardcore ACs and should not be looked down on in any way.
Cheers,
Harry

If you are using RED or still lenses you can use screw on filters.

 

[david farland]

I remember there was a post where Graeme was talking about reducing resolution when you bake in an ASA less than 320.
didn't quite know how this happened , but hey what else is new.
Dave

 

[Tim Sutherland]

Yes, I also remember that post, because to lower noise, I believe the image does some sampling that reduces effective resolution.

Also, if you rate at say 160, I believe you are giving up one stop of dynamic range because highlights are going to blow faster, and you're not simply shifting the DR, but you're pushing one stop over the limit. I really believe that keeping the camera at 360 if at all possible will give you the best results. Period. That's the way the camera was set up. So use ND for bright scenes. I use an 80C or CTB filter for tungsten. Try to add more light instead of rate at 500. The pictures come out much nicer if you do this.

 

[David Mullen ASC]

I'd like some further clarification from RED before I'd believe that lower ASA ratings somehow soften images during conversion from R3D RAW files.

In theory, it's all just metadata, 320 ASA included... by using faster or slower ratings, all you are doing is underexposing or overexposing, biasing your recording for the highlights versus the shadows, trying to find a happy balance between clipping at the one end and shadow noise at the other.

I think you'd get fine results at 250 ASA for example -- that's only 1/3 of a stop more exposure than 320 ASA. Anywhere in that range and you're balancing detail between highlights and shadows fairly equally. Sensors don't really have an ASA rating, just an exposure range in which you fit your information.

The only danger becomes increased clipping as you rate the camera slower and slower. But even that depends on the brightness of highlights in your frame.

Now in terms of the settings for conversion, maybe there is some problems with choosing very low ASA ratings, besides the clipping in the original photography, maybe how it pulls down the brightness is somehow causing some softness, I don't know.

 

[david farland]

I'd love to hear form Graeme on this.
Maybe it's got something to do with when lowering the ASA you 'effectively' increase the -Ev swing of each pixel well, and only having 16 bits (or 12), when you increase the swing you lose granularity on the lower bits. Stab in the dark!!

D

 

[David Mullen ASC]

Again, remember the ASA rating would only be a factor in conversion -- you could rate the camera at 100 ASA... but deliberately underexpose the shot by three stops and then convert the files at 800 ASA setting... In other words, when shooting and recording, all the ASA rating means (other than for monitoring purposes) is that you are biasing your exposure either towards more highlight information or more shadow information. The sensor is what it is in terms of sensitivity and you are just exposing for a certain range of the real world subject.

The question is during conversion if there is some resolution penalty for selecting the lower ASA rating other than the obvious clipping factor that comes into play when overexposing. I mean, let's say you exposed at 100 ASA but opened the files at 320 ASA... and then darkened the footage in a color-corrector. Other than the increase in clipping, would there be a loss of resolution from squashing the brightness of the signal?

 

[david farland]

apologies as I'm new to this but I thought if you plotted the standard -ev count versus brightness level in the Red at it's native ASA (320), you'd have a straight line.
Now I figured when you adjust the ASA you don't actually raise and lower the absolute brightness/shadow levels as you would adding fill, ND's or vary exposure in RedCine.
I thought you added either a convex or concave curve to this linear (at 320) "-ev verses brightness" line.
If you curved this line down (raised ASA), you'd have less bits to represent the variances in the shadows and more representing the hightlights.
The opposite would happen if you curved this transfer line upwards (lowered ASA), so you'd have more of those low end 16 data bits to represent data in the shadows.
I think artifically curving this line to get more data bits describing less actual data (less light) will effect the resolving power (resolution) of system.
D

 

[Daniel Browning]

Hello folks,

Yes, I also remember that post, because to lower noise, I believe the image does some sampling that reduces effective resolution.

I'd like some further clarification from RED before I'd believe that lower ASA ratings somehow soften images during conversion from R3D RAW files.

...all the ASA rating means (other than for monitoring purposes) is that you are biasing your exposure either towards more highlight information or more shadow information. The sensor is what it is in terms of sensitivity and you are just exposing for a certain range of the real world subject.

The question is during conversion if there is some resolution penalty for selecting the lower ASA rating...

I only recall Jim mentioning the resolution difference, with Jared seconding the motion. You can click the "magnifying glass" below to go to the original discussion and context where these quotes were taken from.

The most important quote is the last one.

If you lower the ISO to 100, you will reduce resolution and get a "more sterile" looking image. Not much, but it should be noted.

ISO 100, 125, 160- wicked clean image. Image softens as ISO goes down. Need even more light. Use with care due to resolution decrease.

...you get a resolution hit as you "relax" the RAW past a certain point. Canon has the same issues with their still cameras.

We have found the bug that was causing a slight resolution dip at lower ISOs. There is now no difference.

Hope that helps.

 

[A. Järvinen]

analog vs. digital gain

analog vs. digital gain

I think I understand how red one's iso rating and dynamic range work but I don't quite get it why dlsrs with analog gain have almost the same dynamic range with almost every iso-setting. E.g. according to dpreview.com canon 5dm2 has 8.4-8.6 ev dr at iso 100-3200. I've understood that analog gain (canon 5dm2) and digital gain (red one) are basically the same thing applied in different time, before or after the analog-to-digital-conversion. Could someone please explain why dslrs work the way they do?
Although 8.6 ev is not that much of a dynamic range and it's measured from jpegs, with raw-conversion dpreview.com could get 11.1 ev dynamic range so the image processing has a lot to do with it. But still 11.1 ev - 8.6 ev is 2.5 ev difference and iso 3200 - 100 is 5.0 ev difference. I have thought that the photosite's full-well-capasity determines the dynamic range and the clipping point is a certain level of light, the gain can't change this and this seems to be the way that red one is working but dslrs seem to be somehow different.

(PS. The purpose of this posting is not to compare 5dm2 to red one, just to get clue how different technologies are working. These measurements are from 5dm2's still-pictures, I don't know how it's performing in video and it's irrelevant here. I could have picked any other dslr here for example.)

Best regards,
A. Järvinen

 

[david farland]

...........
1. We have found the bug that was causing a slight resolution dip at lower ISOs. There is now no difference.
2. The absolute best ISO (closest to linear) is ISO 320.......

3. As ISOs increase or decrease from 320, there is a curve adjustment to make them work.
The further away from ISO 320 in either direction, the more the curve adjustment there is and the more "penalty" you pay.
Most recognize that information is "pushed" to accommodate higher ISOs, the converse is true with lower ISOs (mercifully, resolution is not one of them).
While ISO 125 will look great from a noise standpoint, it will not have the the same range as ISO 320 due to the "pulled" curves......

Nice find. Just sound even more interesting.
Am intrigued to know what the 'penalty' is when lowering the ISO.
Also in the thread somone mentioned that 320 was for daylight, but Jim wasn't sure of Red's native ISO for tungsten...(unless they've since measured it).

D

 

[Daniel Browning]

I think I understand how red one's iso rating and dynamic range work but I don't quite get it why dlsrs with analog gain have almost the same dynamic range with almost every iso-setting.

They don't.

E.g. according to dpreview.com canon 5dm2 has 8.4-8.6 ev dr at iso 100-3200.

DPR is a great site, and there is lots of good information, but their stuff about dynamic range is highly flawed. It's a measurement of the tone curve of the JPEG processing engine, nothing more.

I've understood that analog gain (canon 5dm2) and digital gain (red one) are basically the same thing applied in different time, before or after the analog-to-digital-conversion. Could someone please explain why dslrs work the way they do?

Because they have less read noise at high ISO. RED ONE was tested with variable gain controllers and the noise was the same. This is also true in some DSLR. For exmaple, in the 5D2, the noise is the same from ISO 1600 to 3200. However, unlike Canon, RED does not give you the option to use analog gain and throw away dynamic range when there is no improvment in noise. Furthermore, Canon provides three stops of digital ISO (ISO 3200 "pushed" in the camera up to 25,600). RED's digital ISO is metadata and does not delete 3 stops of highlight data for no reason like Canon does. (Canon might have good sensor designers, but their DSLR firmware guys leave something to be desired.)

I have thought that the photosite's full-well-capasity determines the dynamic range and the clipping point is a certain level of light, the gain can't change this and this seems to be the way that red one is working but dslrs seem to be somehow different.

You've got the correct understanding. Skip over the dynamic range section at DPR and you'll be fine.

So, just to summarize, RED's dynamic range is the same all the time since there is no way to change the variable gain amplifier. Here's how dynamic range goes on a DSLR (I'll use the 1D3 for example). This for a fixed exposure (same f-number and shutter speed, but different ISO):

• ISO 100 has the most dynamic range and the least photon shot noise throughout the exposure range. Read noise is the strongest, but it's at the lowest exposure zone, so contributes the least to the total image as long as shadows are not pushed. 11.8 stops of dynamic range using 1:1 SNR per pixel. 7.5 stops of dynamic range using a standard of 4:1 S/N per pixel.
• ISO 200 - one stop of the highest exposure zone is clipped, but read noise is *much* better than ISO 100. Almost enough to make dynamic range the same (more shadow room, less highlight headroom). 11.6 DR @ 1:1 S/N. 6.8 DR @ 4:1 S/N.
• ISO 400 - Read noise is again improved very much, but not as much as the improvement from 100 to 200. 11.3 DR @ 1:1 S/N. 6.0 DR at 4:1 S/N.
• ISO 800 - Same trend. Read noise improved somewhat, but not as much as 200 to 400. 10.8 DR @ 1:1 S/N. 5.2 DR @ 4:1 S/N.
• ISO 1600 - Same trend. 10.1 DR @ 1:1 S/N. 4.2 DR @ 4:1 S/N.
• ISO 3200 - Trend has topped out. No more improvement in read noise. 9.1 DR @ 1:1 S/N. 3.2 DR @ 4:1 S/N.

Those are measured the correct way, with the raw data, from clipping to the the stated read noise level. The 1:1 signal to noise ratio (S/N) is the most common measurement, but I threw in 4:1 S/N just for fun.