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 :)

Dynamic Range

Great…
Any chance of making it down to 100ASA someday?
It would be great in terms of less ND filtering, but would this also mean less noise ?

Emmanuel
 
Emmanuel Cambier said:
Great…
Any chance of making it down to 100ASA someday?
It would be great in terms of less ND filtering, but would this also mean less noise ?

Emmanuel
Click to expand...

Hi Emmanuel,

The sensor is native approx 320 asa your just adding a tag to the raw data.

Stephen
 
You could "pretend" that the ISO is 100, but then you'd run into higlight exposure issues. If you have a very low dynamic range to capture, this might work well and give lower noise still though.

Graeme
 
Ditto to what Stephan said.

There's no point to anything below 320, in fact I don't even know why those options are in there at all, because it's not doing anything useful, the sensor sensitivity isn't changing!
 
If you have a very flat scene, shooting ISO 250 (or even less) is a real choice. Technically, there is less noise at 250 than 320. You just need to be careful you don't clip the highlights. I will say that practically, the noise at 320 is so low that I don't believe it is necessary to do this.

Jim
 
I think I understand it; if you shoot at ISO 100 that just means white clipping will happen 1 2/3 stops sooner than it would at ISO 320, but at the same time you have gained 1 2/3 more stops in your shadows. So it's no substitute for needing ND in the sun, but it seems possible on a low-contrast overcast day, or in mist/fog, etc. But given some red images have been compared with CGI already due to complete absence of visible noise/grain, it doesn't seem that important to reduce noise even more.
 
jbeale said:
I think I understand it; if you shoot at ISO 100 that just means white clipping will happen 1 2/3 stops sooner than it would at ISO 320, but at the same time you have gained 1 2/3 more stops in your shadows. So it's no substitute for needing ND in the sun, but it seems possible on a low-contrast overcast day, or in mist/fog, etc. But given some red images have been compared with CGI already due to complete absence of visible noise/grain, it doesn't seem that important to reduce noise even more.
Click to expand...

Actually you haven't gained any more stops by shooting at 100 ISO. You're just moving more of the image into the low noise better exposed region of the sensor. You only lose latitude one way or the other by adjusting exposure. (After applied in post... of course the normal disclaimers for RAW apply that the data isn't lost... although if you expose for one way or the other on set then you've probably lost detail due to SNR and clipping.)

Turning down the ISO is effectively just reducing the black point so that your shadows start closer to middle exposure and out of the noise floor.
 
im.thatoneguy said:
Actually you haven't gained any more stops by shooting at 100 ISO.
Click to expand...

I definitely agree with you. The overall dynamic range is unchanged. I guess I phrased it poorly. How about: "you loose 1.67 stops in highlights and gain 1.67 stops in shadows", it's a 1:1 tradeoff of what you loose (clip) at the top, you gain back (emerges from noise) at the bottom. (edit: I did make the assumption that you are actually *exposing* for ISO 100, not just adjusting the viewfinder brightness and making a note for the colorist.)

Now IF by chance, you have a scene completely without bright highlights (eg: overcast day exterior) you might still avoid clipping at ISO 100, but I don't know. Maybe if it's a cloudy English countryside and everyone is wearing earth-tone colors, it could work :-). I think the idea is precisely the same as "exposing to the right" on DSLRs. How much highlight clipping you want to expose yourself to (ha) in the quest to optimize your image histogram, is your choice.
 
Just to be clear, the ISO setting on the camera is purely a view setting that affects the monitor outs, plus rides along as metadata with the footage. Changing the ISO setting doesn't alter the recorded image in any way.

Exposing for a low ISO, however, does of course affect the image. If you let more light hit the sensor (by adding light, increasing shutter angle, or opening up the iris on your lens) you can up the signal to noise ratio and see less noise. Trick is to not let so much light in that you're clipping on the upper end.
 
jbeale said:
I definitely agree with you. The overall dynamic range is unchanged. I guess I phrased it poorly. How about: "you loose 1.67 stops in highlights and gain 1.67 stops in shadows", it's a 1:1 tradeoff of what you loose (clip) at the top, you gain back (emerges from noise) at the bottom.
Click to expand...

No that was the part of your statement that was wrong before. The ISO can effectively be thought of as a LUT. However it's a LUT that results in *reduced* dynamic range.

If you opened up one T-Stop on the lens that would shift the entire histogram (losing data to clipping and gaining data in the shadows). If you increase the ISO the white point moves to the left cropping highlights as it goes but the shadows are only increasing in luminance. ISO 50 would probably be equivalent to 8bits of dynamic range.

If somehow the sensor itself could be adjusted (like say... changing film stocks) or some future CMOS chip with variable NDs built in--then you would gain and lose maintaining a 1:1 ratio.

Changing the ISO in post (from RAW) loses 1 stop of Dynamic Range per stop of speed increased or decreased from the native response. If what you were suggesting happened we would actually see detail emerge from the highlights in post as you reduce the ISO. But that's not the case. What has clipped has clipped and what has been lost to noise has been lost to noise. ISO management just crops what's already been captured.
 
Maybe I'm having terminology trouble. When I wrote "shoot at ISO 100" I was thinking that already implied you are exposing for ISO 100. That is, you have consulted your histogram or your light meter, and set your ND/aperture/shutter as you would for ISO 100 film stock. So you are actually capturing more photons per frame. Hence the highlights are 1.67 stops more clipped than had you exposed for ISO 320, but you see 1.67 stops further into the shadows. I think that matches what you're saying. I didn't mean to imply that you could magically recover shadow detail in post if it wasn't already there in the original exposure.

Sekonic-ISO100.jpg
 
Ahhhhh ok I follow now. I thought you were implying that by setting the ISO in the camera to "100" you were actually using a more sensitive sensor. So yes you can expose for 100 iso and get a shot that's a stop too hot just as if you were shooting 5218 rated as 250 asa.

Edit: Now that I think about it the flatter you shoot the scene (say greenscreen studio work) with the camera rated at say 100iso the less noise you're going to get (true) but the more banding you're going to have. So I guess it's really a trade off.

Say you have a scene lit with extremely low contrast (3 stops) with 11 stops of dynamic range that means you would need to underrate the sensor by... ?4 stops? to place that image at the very top of the sensitivity of the sensor. But it also means you're only using about 10bits worth of linear information ((4096*3)/11) which is fine but only about the quarter the color accuracy in exchange for your noise reduced image.

It's an extreme example I know. But shows the trade-offs quite well if you overexpose your image for a 'cleaner' image.
 
This was discussed in some other thread before, but I would like to ask the question to Graeme.
Presume that we have the option in the shutter angle setup menu to alternate shutter angle in such a way that all odd frame will have shutter angle ‘choice #1’ and all even frames shutter angle ‘choice #2’
So if #1 is equal 180 deg and #2 is equal 360 and the speed is 48f/s then we should gain one stop of dynamic range if we down-convert to 24f, is that correct?

If #1=90 and #2=360 then we should get two stops dynamic gain.
Less gain but still significant bust we can get when converting 4K to 2K, however in the above example with alternating shutter speeds we can get all the shadows from the 360 deg frame and combine it in the post with the 90 deg frame that will have all the highlights details in it. Kind of HDR equivalent borrowed from the still photography.

Andrew
 
Theorectically, that works until the camera moves, then you're into the land of trying to temporaly aligning the images and accounting for less motion blur in the shadows.

Graeme
 
It sounds like Gavin is saying that increasing the exposure would increase banding and reduce color accuracy. That doesn't seem right; until clipping occurs increasing exposure always increases the number of bits being used which reduces banding and improves the color accuracy.
 
Temporal alignment will fix itself when averaging two frames from 48f to 24f
After all, our eyes do averaging quite well, on its own.
Averaging may have many forms but even the simplest one work well.
Sure we are not going to use this method for green screen keying.
However to expand dynamic range when needed it would be a great feature.
Until someone will try it we can’t tell how and where we could use it in its best.

I use HDR in still photography since some time and after while I got used to its week points and I do exploit the strong points to its maximum.
I can’t live without it now.
Perfect for slow actions dark or deep shadows shots, or high contrast scenes.
When you need to get out all these shadows in dark or highlights usually you can not have too much movement on the screen anyway.
 
DSPographer said:
It sounds like Gavin is saying that increasing the exposure would increase banding and reduce color accuracy. That doesn't seem right; until clipping occurs increasing exposure always increases the number of bits being used which reduces banding and improves the color accuracy.
Click to expand...

I'm saying avoiding the more noisy shadows by shooting a flatter scene exposed to the right results in less color accuracy.

The flatter the scene the more you can overexpose it without clipping important information. The more you overexpose it (with the intent of increasing the black point to reduce nosie) the less color accuracy you get.

It's true you're still capturing 12 bits of information but if 3/4 of that data is of shadows that you don't want in your image then you've effectively cut your color accuracy by 75%
 
Actually color is never 100% accurate it's just varrying shades of innaccurate. There is near infinite gradation in the real world. You can have no channels clipping but shoot in 4 bit color and have terrible banding. Even 8 bit images will band if the gradation is subtle enough. When grading RED's 12 bit images I've had banding emerge. It all comes down to how far you want to warp the pixels.

Edit:
Pictures are worth a thousand words or so they say:

Our top image here was hypothetically only 6 stops of image that we really wanted (We wanted to crush the shadow details in the trees) and shot with RED's 12 stops of latitude. As a result our DP smartly shot the with it overexposed 2 stops so that all of the detail was as far away from the noise floor as humanely possible without clipping. As a result we have wasted about 50% of the color detail and when expanded (by adjusting the gamma and black point) our overexposed image is about 11 bits worth of color information (or about 50% as much as the original image.)

Seattle_Pano_waveform.jpg
 
I believe more bits means more precision overall, both in color and intensity. And "exposing to the right" means you record more bits (unless you have clipping). If you apply a tone curve you can expand a small range of recorded values and get banding on any camera that exists. But from what I've seen, you can go farther with the Red than with most other cameras out there.

Edit: Thanks for posting the images, although I'm still not sure I get the point. For one thing I have only an 8 bit-per-channel display so if there was color inaccuracy at the 11-bit level, I would not see it. If you are saying that color detail is lost by "exposing right" (without clipping) and then shifting the black point up, as compared with "exposing normally" to achieve the same final desired histogram, then I disagree. It is true that color detail and accuracy may be no better, if your precision is truncated by the bit depth of any intermediate or final output format, but just based on the math it cannot be worse.

I should also note that in the example you show, if the lower image is the desired final output, the whole discussion is moot. That particular exposure could not be taken with any increased exposure without some clipping, because the histogram already extends all the way to the right.
 
You must log in or register to reply here.
Back
Top