Dynamic Range

[Blue]

The crew at Rivetal were capturing 4k files and loading them into Smoke. They were zooming in, color correcting, etc.

It wasn't a side by side comparison between Red and Vision 2 500T. Just an impression from having worked with that film stock and knowing what it looks like in various lighting conditions. There were several there who agreed with me that 1200 ISO looked very similar.

Before my HDV days, that was my film of choice.

I plan to do some actual testing after I get my camera. That should be in about 30 days or less.

I have a client that wants me to show comparisons between Red and several of the popular film stocks, including Vision 2 500T.

I plan to post the results. I think they will be close to my impressions, but we shall see.

-Thor

Cheers mate. Look forward to the results.

 

[Bill Anderson]

Mr Lime, it's tough to say just how much artificial light was available for each of Brasai's night shots. And we're only seeing the successes. Many would be shot at dusk, and not necessarily the dead of night. Nonetheless, the conditions would be pretty challenging. You not only have to consider the film, but the developers that were being used for such conditions. It's amazing just how much you can pull out of a neg with 45 minutes and more developing times with staining developers, Amidol, and waterbath treatments and the likes. Often followed by "intensification" chemistry. These guys were wizards at pulling out the latent image.

 

[Bill Anderson]

Harry, something I should also add, and I'm not sure if this was common practice in Brassai's era, pre-exposing (presensitizing) film to boost its sensitivity under low light conditions is quite effective. The fog produced in the lower values was just printed through. And this method also helped keep in check the added contrast of reciprocity failure. I think this method has even been used in motion picture techniques. Perhaps for different reasons though. Mr Mullen.. ?

 

[David Mullen ASC]

The term "latensification" has been used in motion picture lab work, but it usually just means low levels of flashing or chemical fogging -- true latensification involves very low and long exposure so that the reciprocity effect kicks in, allowing some build-up of photons but not an increase in base fogging. The idea is that it gives barely-exposed silver halide crystals just enough photons to make the crystals "developable".

Flashing can seem to increase shadow detail by lifting the blacks, but there is very little true increase in detail. I remember some tests of the Panaflasher vs. the Varicon by Mark Woods, and one odd thing noticed was that the increase in shadow detail was a little better with the Varicon, which flashes the image through the lens by producing a veil of light over the front of the lens, a controlled glare so to speak. For some reason, the fact that the image was simultaneously flashed, as opposed to pre or post-flashed as with the Panaflasher, boosted the shadow detail.

In really old tube video cameras, there was something called a bias light which had a similar effect in boosting low-level detail, although its main purpose was to reduce image lag.

I ran across an old article in a 1950's issue of "American Cinematographer" which said that Columbia Studios was requiring all movies in the B-Unit to use Plus-X b&w negative using a latensification process to allow them to reduce light levels. I guess this was somehow cheaper for Columbia (which did its own b&w processing back then) than using a faster Kodak b&w stock (like Super-XX or later, Tri-X). But I don't think it was true latensification.

 

[Gregor Hagey]

Toronto Red Test stouffer step wedge tests

Toronto Red Test stouffer step wedge tests

Has anyone done any thorough tests?

I just posted this on exposing for RAW, but perhaps my comments are more relevant here.

My conclusion from the Stouffer chart tests are that processing the .r3d in Red Alert at 320 ISO with -.40 exposure gives you the most latitude. You can apply a slight inverted 'S' curve to pull more out of the shadows and highlights, but the resulting image is very low contrast which is great for exporting for colour grading, but not a good look for editorial. When applying a slight inverted ‘S’ curve by eye I can read 24 steps, but by a waveform monitor 27 steps are definitely visible which equals 9 stops of dynamic range.

Upon analysing the steps on a waveform I noticed that they don’t change on a logarithmic scale. The steps should be in 1/3 stop increments, but at the knee the first two steps are expanded and cover an entire stop each and at the toe the steps are compressed and cover only 1/10 stop. Just like other video cameras there is plenty of information in the shadows, but it’s important to protect your highlights. On a waveform 10 steps (3 1/3 stops) can be seen above 50% and a whopping 17 steps (5 2/3 stops) below 50%. For the purpose of capturing the most dynamic range placing middle grey at 30-35 (instead of 50-55) on your zebras is the best way to go. When I set the camera to 320 ISO (archaically called ASA in the menu) and expose for 30-35% zebras on the viewfinder and then check the T Stop against my incident meter I actually end up with 320 ISO. It seems Red has already built this concept into the ISO settings on the camera. If I want to expose middle grey for 50-55% zebras I need to overexpose by 2 stops. This results in a cleaner image, but you risk clipping the highlights and reducing your overall dynamic range.

So far my real world experience matches these conclusions. You will get into trouble if you underexpose from the ISO you select in the menu (and meter at), but if you go by the zebras and ignore your light meter and put middle grey at 50-55 and skin tone at 60-65 you will have to protect your highlights more, but have no fear of underexposure.

Files can be viewed here and I welcome other DPs evaluation of them:

http://www.dghagey.com/Director_of_Photography/Red.html

 

[Álex Montoya]

Very educative, Gregorios. If the noise floor is indeed improved in the newer firmwares, underexposing a bit looks like indeed the best option.

 

[Graeme Nattress]

As far as I can see, your Stouffer test shows a very log scale, and there is no knee, so I don't know what you're talking about on your chart there. When I take the log of your linear output, I get a straight line, which would mean to me, that the data is linear light and that the stouffer is log, which is correct.

See graphs (yours top, my Stouffer bottom for comparison):
View attachment 1985

Graeme

 

[Gregor Hagey]

I only know how to use a waveform and vectorscope to analyse video and the steps do not progress in a linear fashion at the knee and toe. I consider the dynamic range to begin with the first visible step on the waveform and end with the last step where the video clips.

 

[Graeme Nattress]

Linear light is just that - linear - if you were to put in twice the intensity of light, you'd get a number twice as big. But stops are powers of two - each stop more gives twice the light, so you add 1 stop, but double the light - which is why you get the curve you see - each three steps gets a doubling.

SO, if you take the logarithm on the data, it will plot a totally straight line as shown.

I'm not disagreeing with your measurments - as you point out we all have our own way to define DR. What I am pointing out is that there is no knee or toe in the data - it's a straight line in linear light as my graphs show. Get a copy of ImageJ - very useful for this kind of thing, and do a "log" filter on your linear light image to convert it. Now drag a line selection over your stouffer, and Analyse/plot profile from it. You should get the graph I showed above. I hope that helps!

Remember linear light export exports the image with no curve at all on it. The curve you see in the waveform is because the stouffer is not linear - it's in stops, which are logarithmic. Video is usually gamma encoded, and this is often referred to as "linear", but that's wrong, and bad terminology. It's used because it makes mathematically data "perceptually linear", but that's not a straight line, and of course, linear means a straight line.

So, with linear light data, take a log, plot it - you do get a straight line, as we should for a stop wedge under linear light export.

Graeme

 

[Rudi Herbert]

Gregorios,

Thanks for the detailed report. What are your impressions regarding noise patterns? There's another thread somewhere in the forum where a few users and owners agree that their images pretty much show very bad noise at anything higher than 320 ISO, and that's worrying to me, as I plan to shoot lots of low light exteriors.

Thanks,

 

[David Mullen ASC]

I used to say that grain was a little like salt in your food, no one was going to agree on the proper amount, it was a matter of taste.

Noise is similar -- there is more noise at higher ASA ratings, or with more underexposure, but you'd have to test to determine when it was objectionable for the material you were shooting, and that would also depend on the project's visual needs (dramatically, commercially, etc.) -- and on the planned viewing environment / distribution.

Plus don't forget that for both film and digital, there are noise/grain reduction software. A small amount of noise reduction may be just enough to allow you to use enough underexposure for your particular project.

 

[Gregor Hagey]

Linear light is just that - linear - if you were to put in twice the intensity of light, you'd get a number twice as big. But stops are powers of two - each stop more gives twice the light, so you add 1 stop, but double the light - which is why you get the curve you see - each three steps gets a doubling.

I just re-exported the stouffer as a linear light (the other waveform example was REC 709) and what you say is true, each step doubles in height from the last until it clips. What has me confused is that I usually treat a 10 unit change in height as a doubling in light or a 1 stop change (this doesn't always hold true in the real world, but usually). If each step of the stouffer is a 1/3 stop change in brightness, shouldn't 3 steps line up with each 10 units?

I use Ansel Adams definition of dynamic range "...the first useful values above Zone 0 and below Zone X, or Zones I to IX. The range of zones which convey definite qualities of texture and the recognition of substance is the textural range, from Zones II to VIII." I see Zone 0 as 0 units and Zone X as 100 units. Clearly there's room for interpretation with this definition, but I see at best 24-27 steps of dynamic range. I don't know why this should be considered anything but a fantastic latitude. I've never had this range with any other video camera.

 

[Gregor Hagey]

What are your impressions regarding noise patterns? There's another thread somewhere in the forum where a few users and owners agree that their images pretty much show very bad noise at anything higher than 320 ISO, and that's worrying to me, as I plan to shoot lots of low light exteriors.

Thanks,

I agree with David.

 

[jbeale]

Files can be viewed here and I welcome other DPs evaluation of them: http://www.dghagey.com/Director_of_Photography/Red.html

Wow! I read Reduser quite a bit and had not previously discovered this treasure trove of test data- was it announced before and I just missed it? At any rate, thanks for putting all this online, there's enough here to spend quite a few days looking at. I am curious about the resolution test images, for example http://www.dghagey.com/Director_of_Photography/Lens_Resolution_files/24-S4-75-F5.6.tiff

looking at the center, I see some rather noticeable blue-orange color aliasing. Of course the high contrast chart is sort of a worst-case example, but do you think the anti-alias filter should be stronger?

 

[Graeme Nattress]

I've never seen nasties like that in my 4k zone plate tests, which I think stress the situation to a greater extent than that sharpness chart. I'd need to see the R3D file though. Also, which version of RedAlert was it converted on, as the demosaic has been improved along the way, especially to deal with such issues. To me, looking at the image, it's on an older demosaic than is current, but even for me, it's hard to tell just looking at the pixels. If it's an older REDCINE, then again, newer code which is in sync with REDAlert is much nicer on the red and blue channels.

I feel the OLPF is about right. The extensive aliasing tests I did have always shown very well balanced results - balanced between aliasing and resolution.

Gregor, if you wouldn't mind helping me out on this with a R3D from that resolution shot, it'd be appreciated!

Graeme

 

[Graeme Nattress]

I use Ansel Adams definition of dynamic range "...the first useful values above Zone 0 and below Zone X, or Zones I to IX. The range of zones which convey definite qualities of texture and the recognition of substance is the textural range, from Zones II to VIII." I see Zone 0 as 0 units and Zone X as 100 units. Clearly there's room for interpretation with this definition, but I see at best 24-27 steps of dynamic range. I don't know why this should be considered anything but a fantastic latitude. I've never had this range with any other video camera.

I'm not disagreeing with you at all on this! You're just using a different definition than I am - that's all. That's the thing with DR - we each have our own definition.

I think we can do better than what you saw though - I looked at your stouffer a lot to generate the graphs I've shown, and it didn't seem to extend as far into the shadows as I'd have liked. I pulled up one of my reference stouffers and put it together with yours in Photoshop and it seemed better by a fair bit. Testing DR this way is mighty tricky though, and a lot of factors are involved. Thanks for sharing with us all though, and for the discussion! It prompted me to come up with those new graphs, which I think are pretty interesting, so thanks for that.

Graeme

 

[Gregor Hagey]

Gregor, if you wouldn't mind helping me out on this with a R3D from that resolution shot, it'd be appreciated!

Graeme

Can you PM me where to upload the .r3d? I always use Red Alert for exporting tiffs. It would have been the newest version available the week of October 22.

 

[Daniel Meador]

ok. This question is so basic I am embarrassed to ask. But I have to know anyway. How does HD look on a cinema screen? Can you even project it with a 2K digital projector? Is it practical to spend the extra thousands and wait and wait for a RED SCARLET 3K compared to just buying a Sony PMR EX3 now? My ambition is to screen a film at a film festival. Is HD 'good enough'? And if it can be done what about the 'look'. Audiences are used to seeing film. It's not a 'theatrical movie' in alot of peoples minds unless they see that film look. I mean, people watch video all day long at home on TV. But when they shovel over money they want something different. Film is beautiful if in the right hands. Audiences see it. And what about all the glare and blown out bits video produces. I submit 'Knowing' and 'District 9' as examples. Red digital, Sony, etc............it's all CMOS sensors isn't it?. Now I understnad content is king. The academy award went to a film shot 2k digital.

Jannard has made alot of claims so far about the amazing progress he is making. Of course we are all waiting to see it. All we know now is that the hardware is pleasingly estetic Do I wait and take it all on faith? I'd like to know what a professional cinematographer such as Mr. Mullins thinks as well. It's hard to ride this wave of bleeding edge digital technology for months on end. I want to make ART now. But I don't want to waste money on an HD camera or for that matter 'todays technology' ie RED ONE, if on the near horizon is the SECOND COMING. Know what I mean?

 

[geoffboyle]

I missed this thread when I originally posted about the Hannover tests.

I continue to test the way that I've always tested as this gives me the chance to compare tests from now with those of 25 years ago, yes I have been testing longer than that.

I have put examples of all the tests on the CML website, you don't have to be a member to access any of these.

You can also be a member to get access to the members messages directly without getting ANY emails, there's an option in the user setting for web only. It's been there for 9 years.

Initially we were only email but that was '96 to '00

As long as you know how test were done, and I always detail this, you can learn a lot from them.

I haven't published any of my most recent tests yet because I have found what I consider to be inconsistencies and I'm talking to RED about this before I publish, I'm also talking to Technicolor.

I'm also looking at the effect of different de-bayering processes.

 

[Dan Hudgins]

Preamps

Preamps

looking at the center, I see some rather noticeable blue-orange color aliasing. Of course the high contrast chart is sort of a worst-case example, but do you think the anti-alias filter should be stronger?

On the center pattern there seems to be an asymmetry in the vert and horz resolution, I seem to see that in other sensors as well and have been looking into the cause.

The first thing that seems logical would be the pre-amp bandwidth before the ADC, since before they have posted sine wave charts used for adjusting the so called 3.2K resolution, if the horz resolution is lower than the vert resolution then if they adjust for the horz the vert will be peaked too much.

To compensate for that you need to blur the image in the vert to make the resolutions equal? (or use better preamps)

=

On the issue of Dynamic range, it would be good if the code for REDCINE and REDALERT would have a "optical balance" setting for their de-Bayer, that way you can get the full 12bits of dynamic range, more or less.

Because the sensor is not 100% in blanace with any standard color of light some clipping is done to balance the white clip levels for R,G, and B.

Since there is not adjustment for the clip levels that gives total fine control in REDCINE and REDALERT you need to add an option to set those clip levels at least to 100% signal, then an optical filter can be used to balance the sensor to improve the dynamic range.

This is a simple feature, and its hard to understand why it has not been included all along?