Let's talk about Dynamic Range at DVinfo.net

Kris Bird · October 23rd, 2007, 09:36 AM

I've been able to find very little on the subject in the forums... And I don't mean DR theory, I mean very specific tweaking of custom presets with the intent on controlling/maximising DR.

I'm currently working on finalising my custom presets for our feature length fiction, and I've spent some time with charts and the Canon Console +waveform. I'm *not* shooting high spec perfectly lit charts, just a variety, some home made, some still-life setups.

One thing I see mentioned in the forum search is that the XH-A1 has a wider dynamic range when set to 'video' gamma, rather than cine1 or cine2. This doesn't match my findings at all, am I missing something? As far as I can see, the gamma makes the curve more or less steep (obviously), effectively changing the priority/allocation of bits, but Video gamma certainly isn't expanding anything. If anything, Video gamma compresses the DR of the highs (even with low knee). Sure, this is to free up bits further down the scale (why we're applying gamma in the first place) and results in a nice bright video image, but I see less steps at the top end and basically the same at the bottom.

Does anyone know what settings Adam Wilt used when he measured 8.3 stops of DR? Obviously this is >8stops which we classically say can be decently represented by 8 bits, so in theory we're on the verge of subtle banding somewhere in the scale, but I'd like to know what he used.

If there's anything I'm missing please let me know.

What are other people's findings?

Kris

Juan Diaz · October 23rd, 2007, 09:51 AM

Hi Kris,

Using one of the Cine gammas should give you more dynamic range than the standard gamma. I believe Cine2 should give you more than Cine1 because it's designed for doing a filmout (as opposed to the "look of film on video").

Also, you should set Black=Stretch and Knee=Low for maximum dynamic range. These settings will result in a somewhat low contrast look, of course, but you can always push contrast in post.

Kris Bird · October 23rd, 2007, 10:06 AM

Cheers Juan, that's exactly my findings.. I thought I was going mad. My logic at the moment is that Cine2 gives me the most range, based on visually counting steps and trying every which combination of settings (along with black stretch and knee low as you said), but at the theoretical expense of having a large portion of the image quite 'dark', and therefore needing some boost in post, and subsequent risk of increased banding in the lows. I've also got Master Ped at -4, to get it as low as possible without low-clipping the very bottom of the signal, to make sure I'm using all the bits I can.

So is this what other people have found? re: use of cinegamma specifically.

Brandon Freeman · October 23rd, 2007, 06:22 PM

Having no idea what the master pedestal did, I had been keeping video gamma levels at normal with blacks stretched and highlights crushed. I was happy with the amount of range I got and accepted that I would just have to lower the curve a bit in post in order to get the look I wanted.

But, after reading this thread, I decided to look at CineGamma 2 again. I liked what it did, as I always have, but I felt it made the picture look too dark, and was actually eliminating dynamic range as opposed to increasing it, because I would have to open up or gain up to get dark objects back to the same exposure as before, and lighter objects would blow out.

Then...I played with the master pedestal... And amazingly, it brought the exposure on the bottom end back up while still leaving the gamma tweak in full effect. I'm not sure what the curve actually looks like, but the image is amazing.

BLACK STRETCH
KNEE LOW
CINEGAMMA 2
MASTER PED AT +9
CINE COLOR MATRIX 2

Kris Bird · October 24th, 2007, 08:58 AM

Brandon,

That is what I'm doing, but raising Master Ped isn't needed! If you look at curves/console while pointing at a scene, you'll see that Master Ped moves the very bottom of the curve up and down, but doesn't EXPAND it ... so your values are scaled up from say 8-255, to say 20-255. This makes the dark things LOOK brighter, but it isn't "unclipping" anything .... Well, I should say- it may raise details out of the pitch black when watching back on your *TV*, but if you're colour correcting in post then it isn't giving your any more information that you can't get back with curves. In fact, all you're doing is wasting some bits, as you're using less of the full range. I.e. you're not making use of the bit of the scale from "REALLY BLACK" to "BLACK", all look black on your TV, but if you're grading in post you can use that ..... My in progress preset is using -4 master ped, it's as low as it can go with my current settings before the scene bottom-clips.

Brandon Freeman · October 24th, 2007, 09:46 AM

Kris, that's entirely true, and most of the stuff I shoot I have to darken on the low end in post, but I find that I get better results if I crunch in post as opposed to stretch. It seems to result in less noise. I'll do some more testing.

Kris Bird · October 24th, 2007, 10:21 AM

Brandon, I agree, I've always done it that way... shoot high and push contrast in post ... But what if a project is going to be relatively dark anyway, and cine1 or cine2 are closer to your end result than video? In the past I've assumed (and read) that by doing this we'd lose dynamic range and all our detail in the darks... but it doesn't seem to be true here.

This is what I'm toying with..

1. How to maximize measurable dynamic range, even if it necessitates post-fixing

2. Whether this post-fixing (for example lifting things back up) is acceptable, or even beneficial in some cases (where you make use of super blacks and super whites, for example, and fix in a 10bit+ post workflow to avoid/limit rounding errors).

3. Whether one gamma mode is demontratably 'best' (in a wide range of scenarios), or whether even varying gamma mode by scene depending on desired content/result is actually the best use of bits. ie. if it's meant to be a low key scene, shoot cine2 and make the best use of bits, if it's meant to be higher key use video gamma.

*please* get in touch (anyone) if you've differing results/conclusions, take none of this as gospel

Kris Bird · October 24th, 2007, 07:58 PM

More testing, subjective finding--

1. When grading two captured videos of an identical scene, one clip Cine2 Gamma one clip Video Gamma, I vastly prefer the grading result I get with the Cine2 Gamma. I need less funky curve adjustment in order to get the right texture/contrast I'm after. By 'funky' I mean that I'm not having to use 5-6 points on a curve graph, I can get what I'm after with two points. I find that really interesting, it's a good development. If you find yourself with finicky curves trying to 'uncompress' some contrast back into your highlights, and a little bump at the very bottom to keep your lows visible before a dip in the lower/mid area to add contrast, etc., then I seriously recommend you check out cine2, it has a great aesthetic, *presuming* that you're already shooting and exposing something decent.

2. You CAN match Cine2/Video captures pretty identically, with a slightly funky Curve adjustment and saturation adjustment, at which point the dynamic range in both images looks really very close. So as I mentioned before, a lot of this might just be relating to getting as close as possible to your target aesthetic as possible, to make the most out of your 8 bits and not be re-inventing the wheel with crazy curves in the grade.

.. still would love to hear the results of people's experiments

Jury's still out on whether/ how much gamma affects dynamic range, beyond the (perhaps recoverable) compression of the highlights.

Brandon Freeman · November 6th, 2007, 01:20 PM

Ivan Mosny · November 6th, 2007, 02:24 PM

Dynamic range in Picture is a difference between full white and full black points.
Video gamma is linear. Cine gamma curves are in form of "S". Cine1, Cine 2 or Video - all of them have the same begin and end points = dynamic range is always the same. Cine gammas makes even dark parts darker with less differences between dark areas (but not darker as a full black point) and brighter areas brighter (but not brighter as full white). You can do the same in postproduction. Cinegamma makes "Filmlook" but not a film dynamic range. Dynamic range depends on camera picture sensors - not on internal gamma processing. Remember - cinegamma looks good - but its only electronically distorted video response. Its not real life.

Kris Bird · November 6th, 2007, 03:39 PM

Ivan-

A cine gamma curve isn't an "s" contrast curve, video gamma isn't "linear" .. in fact it's kind of the opposite. Cine gamma is less 'gamma corrected' in a traditional video sense, i.e. it is closer to the real life linear response. CCDs do operate in the real world. The curve is shaped as per gamma's power law relationship-

http://www.dfanning.com/ip_tips/xstretch_5.jpg

Lower gamma means that your image is perceptually dark at the encoding/quantisation stage, resulting in a re-biasing of the available 8bits. Your image comes into post looking dark/flat. What you describe (the 's' contrast boost) is endemic to 'video' gamma- the lows and highs are compressed/crushed, to give a perceived contrasty image. This crushing means basically what you describe- darks get closer to black, whites get closer to white... while white and black do stay where they are, the 'usable', recoverable information is often lost, or practically lose (i.e. death in quantisation). Remember also that these curves are applied before they are fully mapped/quantised/etc in camera, i.e. before the white point and black point are chosen (note- setup, master ped, etc.), so dynamic range 'can' indeed be affected.

-EDIT-
Here you go- http://www.usa.canon.com/app/html/HD...justment.shtml

Kris Bird · November 6th, 2007, 03:41 PM

Brandon- cool, nice test shots, looking at them now.

The interior is easy to see- there's some bottom-clipping in the darkest bit of the curtain about 3/4 of the way across (from left), this is totally in tact right down to say -4 (i.e. you may as well shoot down here at all times), with the slightest hint of compression at -5.. getting bad at -7. To be fair, you could probably argue that the subtle extra bit-precision (for the whole image) outweighs the subtle extra DR for a tiny bit of curtain shadow detail in this case, right down to even -7 or -8. Really depends how much of your image is going to be scraping the bottom of the DR barrel..! In terms of shooting with a higher master ped to help with monitoring (on set, or tv, etc.), you probably aren't losing much. Looking at the outside image, there's almost no bottom-clipping .. pretty low DR scene interestingly (99% within the A1s lattitude). I see a hint of compression starting about 1/3 of the way from the left of image, but hard to make it out until -7. Are you seeing the same? Seems like we can safely say that (for setup=0, black=stretch, gamma=cine2, at least) MP > -4 is only really for when you need your output bright for viewing purposes, lower is better for best use of bits, ~-4 is as low as you can go without seeing even a hint of compression of near-blacks, beyond that sees some low-compression but can be entirely acceptable for a relatively bright scene, and makes best use of bits.

Ivan Mosny · November 6th, 2007, 04:45 PM

Quote:

Originally Posted by Kris Bird

Remember also that these curves are applied before they are fully mapped/quantised/etc in camera, i.e. before the white point and black point are chosen

I don`t think so. Have you any source for this Information?
Black and White levels on Video signal are stable levels - independent from gamma settings or character of picture (in both - digital or analog video).
IMHO this Cinegamma correction is aplicated to compelete videosignal in a input/output matrix. But okay - am not a Canon constructor - i can not be sure.

Kris Bird · November 6th, 2007, 04:52 PM

Quote:

Originally Posted by Ivan Mosny

I don`t think so. Have you any source for this Information?
Black and White levels on Video signal are stable levels - independent from gamma settings or character of picture (in both - digital or analog video).
IMHO this Cinegamma correction is aplicated to compelete videosignal in a input/output matrix. But okay - am not a Canon constructor - i can not be sure.

If you alter Setup or Master Ped then the whole curve shifts up and down, so the gamma correction does happen before that in the image chain... I'm not saying this is a major factor, just that black and white aren't absolutes.

Kris Bird · November 6th, 2007, 05:06 PM

Also if you point the cam at a ramp which overexposes at the top 20%, then set zebras to 100%, then changing the gamma curve changes the point where the zebras start. I haven't yet tested whether this is because the 0.0-1.0 points for the gamma transform are different to the 0.0-1.0 for the output (i.e. internal white is higher than output white to limit DR due to 8bit/contouring), or whether it is simply that 100% zebras are broadcast legal limit, rather than output 1.0.

October 23rd, 2007, 09:36 AM	#1
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	Let's talk about Dynamic Range I've been able to find very little on the subject in the forums... And I don't mean DR theory, I mean very specific tweaking of custom presets with the intent on controlling/maximising DR. I'm currently working on finalising my custom presets for our feature length fiction, and I've spent some time with charts and the Canon Console +waveform. I'm not shooting high spec perfectly lit charts, just a variety, some home made, some still-life setups. One thing I see mentioned in the forum search is that the XH-A1 has a wider dynamic range when set to 'video' gamma, rather than cine1 or cine2. This doesn't match my findings at all, am I missing something? As far as I can see, the gamma makes the curve more or less steep (obviously), effectively changing the priority/allocation of bits, but Video gamma certainly isn't expanding anything. If anything, Video gamma compresses the DR of the highs (even with low knee). Sure, this is to free up bits further down the scale (why we're applying gamma in the first place) and results in a nice bright video image, but I see less steps at the top end and basically the same at the bottom. Does anyone know what settings Adam Wilt used when he measured 8.3 stops of DR? Obviously this is >8stops which we classically say can be decently represented by 8 bits, so in theory we're on the verge of subtle banding somewhere in the scale, but I'd like to know what he used. If there's anything I'm missing please let me know. What are other people's findings? Kris

October 23rd, 2007, 06:22 PM	#4
Brandon Freeman Major Player Join Date: Feb 2007 Location: Boise, ID Posts: 226	Having no idea what the master pedestal did, I had been keeping video gamma levels at normal with blacks stretched and highlights crushed. I was happy with the amount of range I got and accepted that I would just have to lower the curve a bit in post in order to get the look I wanted. But, after reading this thread, I decided to look at CineGamma 2 again. I liked what it did, as I always have, but I felt it made the picture look too dark, and was actually eliminating dynamic range as opposed to increasing it, because I would have to open up or gain up to get dark objects back to the same exposure as before, and lighter objects would blow out. Then...I played with the master pedestal... And amazingly, it brought the exposure on the bottom end back up while still leaving the gamma tweak in full effect. I'm not sure what the curve actually looks like, but the image is amazing. BLACK STRETCH KNEE LOW CINEGAMMA 2 MASTER PED AT +9 CINE COLOR MATRIX 2 __________________ Canon EOS 7D, Final Cut Studio 2

October 24th, 2007, 09:46 AM	#6
Brandon Freeman Major Player Join Date: Feb 2007 Location: Boise, ID Posts: 226	Kris, that's entirely true, and most of the stuff I shoot I have to darken on the low end in post, but I find that I get better results if I crunch in post as opposed to stretch. It seems to result in less noise. I'll do some more testing. __________________ Canon EOS 7D, Final Cut Studio 2

October 24th, 2007, 10:21 AM	#7
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	Brandon, I agree, I've always done it that way... shoot high and push contrast in post ... But what if a project is going to be relatively dark anyway, and cine1 or cine2 are closer to your end result than video? In the past I've assumed (and read) that by doing this we'd lose dynamic range and all our detail in the darks... but it doesn't seem to be true here. This is what I'm toying with.. 1. How to maximize measurable dynamic range, even if it necessitates post-fixing 2. Whether this post-fixing (for example lifting things back up) is acceptable, or even beneficial in some cases (where you make use of super blacks and super whites, for example, and fix in a 10bit+ post workflow to avoid/limit rounding errors). 3. Whether one gamma mode is demontratably 'best' (in a wide range of scenarios), or whether even varying gamma mode by scene depending on desired content/result is actually the best use of bits. ie. if it's meant to be a low key scene, shoot cine2 and make the best use of bits, if it's meant to be higher key use video gamma. please get in touch (anyone) if you've differing results/conclusions, take none of this as gospel Last edited by Kris Bird; October 24th, 2007 at 11:42 AM.

November 6th, 2007, 01:20 PM	#9
Brandon Freeman Major Player Join Date: Feb 2007 Location: Boise, ID Posts: 226	Okay, so I've heard from a few different sources now that I'm wasting bits having the Master Pedestal amped up, so, how about in the other direction? Blacks stretched, knee low, CineGamma 2, Cine Color Matrix 2. I did a test with the pedestal from 0 to -9, both indoors and outdoors -- let me know what you think. Pedestal - Indoors http://www.hall-e-woode.com/test_pic...pedestal_0.jpg http://www.hall-e-woode.com/test_pic...edestal_-1.jpg http://www.hall-e-woode.com/test_pic...edestal_-2.jpg http://www.hall-e-woode.com/test_pic...edestal_-3.jpg http://www.hall-e-woode.com/test_pic...edestal_-4.jpg http://www.hall-e-woode.com/test_pic...edestal_-5.jpg http://www.hall-e-woode.com/test_pic...edestal_-6.jpg http://www.hall-e-woode.com/test_pic...edestal_-7.jpg http://www.hall-e-woode.com/test_pic...edestal_-8.jpg http://www.hall-e-woode.com/test_pic...edestal_-9.jpg Pedestal - Outdoors http://www.hall-e-woode.com/test_pic...pedestal_0.jpg http://www.hall-e-woode.com/test_pic...edestal_-1.jpg http://www.hall-e-woode.com/test_pic...edestal_-2.jpg http://www.hall-e-woode.com/test_pic...edestal_-3.jpg http://www.hall-e-woode.com/test_pic...edestal_-4.jpg http://www.hall-e-woode.com/test_pic...edestal_-5.jpg http://www.hall-e-woode.com/test_pic...edestal_-6.jpg http://www.hall-e-woode.com/test_pic...edestal_-7.jpg http://www.hall-e-woode.com/test_pic...edestal_-8.jpg http://www.hall-e-woode.com/test_pic...edestal_-9.jpg __________________ Canon EOS 7D, Final Cut Studio 2

October 23rd, 2007, 09:51 AM	#2
Juan Diaz Regular Crew Join Date: Oct 2005 Location: Austin, TX Posts: 49	Hi Kris, Using one of the Cine gammas should give you more dynamic range than the standard gamma. I believe Cine2 should give you more than Cine1 because it's designed for doing a filmout (as opposed to the "look of film on video"). Also, you should set Black=Stretch and Knee=Low for maximum dynamic range. These settings will result in a somewhat low contrast look, of course, but you can always push contrast in post.

October 23rd, 2007, 10:06 AM	#3
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	Cheers Juan, that's exactly my findings.. I thought I was going mad. My logic at the moment is that Cine2 gives me the most range, based on visually counting steps and trying every which combination of settings (along with black stretch and knee low as you said), but at the theoretical expense of having a large portion of the image quite 'dark', and therefore needing some boost in post, and subsequent risk of increased banding in the lows. I've also got Master Ped at -4, to get it as low as possible without low-clipping the very bottom of the signal, to make sure I'm using all the bits I can. So is this what other people have found? re: use of cinegamma specifically.

October 24th, 2007, 08:58 AM	#5
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	Brandon, That is what I'm doing, but raising Master Ped isn't needed! If you look at curves/console while pointing at a scene, you'll see that Master Ped moves the very bottom of the curve up and down, but doesn't EXPAND it ... so your values are scaled up from say 8-255, to say 20-255. This makes the dark things LOOK brighter, but it isn't "unclipping" anything .... Well, I should say- it may raise details out of the pitch black when watching back on your TV, but if you're colour correcting in post then it isn't giving your any more information that you can't get back with curves. In fact, all you're doing is wasting some bits, as you're using less of the full range. I.e. you're not making use of the bit of the scale from "REALLY BLACK" to "BLACK", all look black on your TV, but if you're grading in post you can use that ..... My in progress preset is using -4 master ped, it's as low as it can go with my current settings before the scene bottom-clips.

October 24th, 2007, 07:58 PM	#8
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	More testing, subjective finding-- 1. When grading two captured videos of an identical scene, one clip Cine2 Gamma one clip Video Gamma, I vastly prefer the grading result I get with the Cine2 Gamma. I need less funky curve adjustment in order to get the right texture/contrast I'm after. By 'funky' I mean that I'm not having to use 5-6 points on a curve graph, I can get what I'm after with two points. I find that really interesting, it's a good development. If you find yourself with finicky curves trying to 'uncompress' some contrast back into your highlights, and a little bump at the very bottom to keep your lows visible before a dip in the lower/mid area to add contrast, etc., then I seriously recommend you check out cine2, it has a great aesthetic, presuming that you're already shooting and exposing something decent. 2. You CAN match Cine2/Video captures pretty identically, with a slightly funky Curve adjustment and saturation adjustment, at which point the dynamic range in both images looks really very close. So as I mentioned before, a lot of this might just be relating to getting as close as possible to your target aesthetic as possible, to make the most out of your 8 bits and not be re-inventing the wheel with crazy curves in the grade. .. still would love to hear the results of people's experiments Jury's still out on whether/ how much gamma affects dynamic range, beyond the (perhaps recoverable) compression of the highlights.

November 6th, 2007, 02:24 PM	#10
Ivan Mosny Regular Crew Join Date: Sep 2007 Location: Poland Posts: 141	Dynamic range in Picture is a difference between full white and full black points. Video gamma is linear. Cine gamma curves are in form of "S". Cine1, Cine 2 or Video - all of them have the same begin and end points = dynamic range is always the same. Cine gammas makes even dark parts darker with less differences between dark areas (but not darker as a full black point) and brighter areas brighter (but not brighter as full white). You can do the same in postproduction. Cinegamma makes "Filmlook" but not a film dynamic range. Dynamic range depends on camera picture sensors - not on internal gamma processing. Remember - cinegamma looks good - but its only electronically distorted video response. Its not real life.

November 6th, 2007, 03:39 PM	#11
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	Ivan- A cine gamma curve isn't an "s" contrast curve, video gamma isn't "linear" .. in fact it's kind of the opposite. Cine gamma is less 'gamma corrected' in a traditional video sense, i.e. it is closer to the real life linear response. CCDs do operate in the real world. The curve is shaped as per gamma's power law relationship- http://www.dfanning.com/ip_tips/xstretch_5.jpg Lower gamma means that your image is perceptually dark at the encoding/quantisation stage, resulting in a re-biasing of the available 8bits. Your image comes into post looking dark/flat. What you describe (the 's' contrast boost) is endemic to 'video' gamma- the lows and highs are compressed/crushed, to give a perceived contrasty image. This crushing means basically what you describe- darks get closer to black, whites get closer to white... while white and black do stay where they are, the 'usable', recoverable information is often lost, or practically lose (i.e. death in quantisation). Remember also that these curves are applied before they are fully mapped/quantised/etc in camera, i.e. before the white point and black point are chosen (note- setup, master ped, etc.), so dynamic range 'can' indeed be affected. -EDIT- Here you go- http://www.usa.canon.com/app/html/HD...justment.shtml Last edited by Kris Bird; November 6th, 2007 at 04:48 PM.

November 6th, 2007, 03:41 PM	#12
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	Brandon- cool, nice test shots, looking at them now. The interior is easy to see- there's some bottom-clipping in the darkest bit of the curtain about 3/4 of the way across (from left), this is totally in tact right down to say -4 (i.e. you may as well shoot down here at all times), with the slightest hint of compression at -5.. getting bad at -7. To be fair, you could probably argue that the subtle extra bit-precision (for the whole image) outweighs the subtle extra DR for a tiny bit of curtain shadow detail in this case, right down to even -7 or -8. Really depends how much of your image is going to be scraping the bottom of the DR barrel..! In terms of shooting with a higher master ped to help with monitoring (on set, or tv, etc.), you probably aren't losing much. Looking at the outside image, there's almost no bottom-clipping .. pretty low DR scene interestingly (99% within the A1s lattitude). I see a hint of compression starting about 1/3 of the way from the left of image, but hard to make it out until -7. Are you seeing the same? Seems like we can safely say that (for setup=0, black=stretch, gamma=cine2, at least) MP > -4 is only really for when you need your output bright for viewing purposes, lower is better for best use of bits, ~-4 is as low as you can go without seeing even a hint of compression of near-blacks, beyond that sees some low-compression but can be entirely acceptable for a relatively bright scene, and makes best use of bits. Last edited by Kris Bird; November 6th, 2007 at 04:27 PM.

November 6th, 2007, 05:06 PM	#15
Kris Bird Regular Crew Join Date: Sep 2005 Location: UK Posts: 173	Also if you point the cam at a ramp which overexposes at the top 20%, then set zebras to 100%, then changing the gamma curve changes the point where the zebras start. I haven't yet tested whether this is because the 0.0-1.0 points for the gamma transform are different to the 0.0-1.0 for the output (i.e. internal white is higher than output white to limit DR due to 8bit/contouring), or whether it is simply that 100% zebras are broadcast legal limit, rather than output 1.0.