Avoiding Noise in the Shadows

[Esteban Sosnitsky]

Hi all, I just asked this in the ASK DAVID MULLEN anything thread, but would like an opinion from all of you aswell as it probably isnt just about what the DP has to do but what the DIT or post processor has to do.
:help:
How does one produce low light images without generating noise?
I have made a few exercises and there must be something I am doing wrong because I always tend to see noise in the shadows. Can someone tell me how to produce images on red without noise in the dark areas?
Is it on camera?
Is it because I still have the build 15?
Does it depend on post processing?
Does it depend on the settings in redalert redcine?
Does it depend on the color gamma spaces?
I feel a little disappointed, I have tried very different settings and I always get noise in the shadows. Any suggestions on what to experiment would be very much appreciated. Thanks!
Thanks.

[Fredrik Callinggard]

In build 15 blue channel noise in low light was, for me and others, quite well.... noisy (this is a matter of preference, of course). Especially in low light tungsten conditions. What was worse was that the noise had a compression artifact, which made it almost untreatable.

That has all changed with 16 :biggrin:

[Esteban Sosnitsky]

Thanks for your response!
I downloaded some footage processed by someone in the forum and they were almost noise free images in low light situations. Looking forward to see if anyone who has done that knows and wants to share.
Thanks again Fredrik.

[Mike Prevette]

One thing to note is the RED doesn't roll off the noise floor like almost every other camera. This is good because the camera manipulates the raw data less, and you get to see EVERYTHING the camera captures. BUT It takes a while to get used to having to roll off the bottom of the curve before it's usable.

Also processing with the highest quality debayer will help, as will moving to Build 16.

[Billy Summers]

BUT It takes a while to get used to having to roll off the bottom of the curve before it's usable. .

Rolling off the bottom of the curve..."in Post"

[Esteban Sosnitsky]

Thanks again for your replies. But I really dont understand what rolling off the bottom of the curve means. Also, I did some tests and I see a lot of noise in grey and blue areas of the image. Is this normal? Is this what they said is solved in B16?

[David Mullen ASC]

It's pretty simple -- underexposed information falls further and further into a zone of noise at the bottom near black. You want less noise, then expose so that the information you don't want to be noisy is higher than the noise level.

The only risk of doing that is the other end of the scale, the clip point -- you slide your exposure information further up the scale... and push more overexposure information off the cliff edge of clipping.

So you want less noise, expose more and control the bright highlights that you don't want to clip through lighting, using darker colors in wardrobe, walls, etc. Then the shadow information will be less noisy and you can crush the very bottom of the scale into black and get a very clean image.

That or use noise reduction processing at the risk of some softening, perhaps smearing of motion.

[Joel Kaye]

How does one produce low light images without generating noise?

The truth as I see it is you really can't underexpose RED much if at all in low light. Which is to say, if your light meter reads ISO 320 F1.4 and you shoot at F2.8 you're going to get some ugly RED. But if you shoot at 1.4 it'll look pretty good. So sunrise, sunset where you have a pretty even 1.4 at 320 you get some nice low noise stuff even though it seems like low light. It is, but the exposure is correct. Being daylight instead of tungsten seem to help too.

I still think the camera is a 160 camera but underexposes to 320 perfectly. Go to 800 in very low tungsten light and you're asking for trouble. That has never worked great for me. It's just noisy. At least the noise pattern in 16 is more organic so you can just go with it. Scaling all the way down to SD helps make it look better.

[Jason Sinclair]

I get the feeling you need more basic information? Correct me if i'm wrong.
You are probably exposing by eye. You are looking at the image in low light and you are increasing either exposure or ISO to get the image brighter and then you get the noise? If this is the case then you need to drop the exposure or iso down. If you find this creates too many shadows then you need to bring in more light. Unfortunately, really, that is the nature of image production. It is true for film as much as it is for digital.

[David Mullen ASC]

If someone said "I'm getting too much grain in the 500T emulsion I'm using" -- my response would be: (1) overexpose the stock more, or (2) use a slower-speed stock, or (3) use a bigger negative format.

There's no free lunch -- too little exposure produces grain on film or noise in digital. If you are getting too much noise in the shadows, you need more exposure in the shadows. More exposure gives you the flexibility of darkening the image in post without getting more noise, whereas too little exposure not only puts more shadow detail in the noise floor, but any color-correcting in post that plays around with the shadows, particularly any attempt to lighten them, only brings up the noise.

The only reason that 3200K is noisier than 5600K is that when your lighting is warmer in color, you are basically underexposing the blue channel more and more, which is why it gets noisier when you attempt to correct that low color temperature light to "white" by pushing the underexposed blue channel.

[Esteban Sosnitsky]

Ok, thanks for your input, its really appreciated.
David, I do understand what you are saying, the thing is I have tested the camera with daylight and tungsten in different situations and I feel the camera noise is too high. Probably with B16 this issue will be different BUT I saw Gunleik's TESTS footage and I dont see that kind of noise there. I wander if he did some post effects or something, looking forward to his response.
I exposed by eye with a monitor (which is not good because of color space and due the fact that the output is 720p) but I always check with my meter to be sure and make a decision from there. I have worked with the F900 and the F23 and I dont see this kind of effects there.
I feel your input is very valuable in the means that exposing the shadows a little more will get you more info to pull back afterwords. Thanks. But for example, the other day I shot a demo which had a blue background. I lit it with daylight HMI;s and KINO flos, and the amount of noise in the whole picture was a little too high. I have an upcoming shoot which should be very contrasty. I guess I will light it flat and contrast in post? What concerns me about this means of procedure is the clipping highlights, so I will try to be more accurate on the exposure and comment on this thread afterwords.;)

Thank you all for your help. I still would like to hear if someone is doing something different in post, other than what we have discussed here. Thanks again.

[Esteban Sosnitsky]

I still think the camera is a 160 camera but underexposes to 320 perfectly. Go to 800 in very low tungsten light and you're asking for trouble. That has never worked great for me.

Hi Joe, I agree, I have never used the camera higher than iso 500. I tend to use 320 and either dim the lights or go to 160. The thing is sometimes, in low budget productions your lights can only be tungsten cost wise. So I am trying to experiment in both situations and see where can I get this camera to work nicely under these circumstances. I just shot a greenscreen with tungsten, rated the camera at 4000K (WB) and iso 320. I will post the results, I can say they look great. But I want you to see it.;)
Nevertheless, whenever the light falls under zone IV I feel the noise is not good in either daylight or tungsten, of course I see it more in tungsten because of what david explained earlier, but I have to be fair and I havent seen that kind of noise in the Cinealtas. (Dont judge me, I bought a RED;))

[Esteban Sosnitsky]

I get the feeling you need more basic information? Correct me if i'm wrong.
You are probably exposing by eye. You are looking at the image in low light and you are increasing either exposure or ISO to get the image brighter and then you get the noise? If this is the case then you need to drop the exposure or iso down. If you find this creates too many shadows then you need to bring in more light. Unfortunately, really, that is the nature of image production. It is true for film as much as it is for digital.

Supernova, I expose with the meter and by looking at the monitor and the scene itself. I never increase the ISO, all my shoots are 160, 320 or 500 tops in worst case scenarios. Thanks for your input.:wink:

[David Mullen ASC]

Maybe you could post some example frames of the noise problems you are encountering?

[Esteban Sosnitsky]

I will try, as I shot this for commercial use, I would have to ask.

[Jason Sinclair]

What kind of meter you using?

Expose for the highlights/gamma.
Do tests on your hmi's...
What monitor are you using?
Try tests using full stop range and bring into studio and find where noise stops. Find the discrepancy with your light meter readings and adjust accordingly.

If you are using just the small lcd and light meter, then i would say this is a light metering problem.

I mean if you see the noise on monitor why don't you just drop exposure???

Or does noise still occur when you do this?

If so then yes please post frame grabs...

[Jason Sinclair]

I will try, as I shot this for commercial use, I would have to ask.

just the noise will be ok. Some area of shadows/highlights. We don't need to see the whole frame.

[Esteban Sosnitsky]

Just finished another shoot. I applied what you have suggested and I have to say I feel the problem has been minimized. I over exposed a stop of what I would have decided previously and then pull down the curve in red alert maximizing contrast, deep blacks, rich mid tones and highlight detail are present. I feel the problem was that nailing the exposure is really hard when the monitor looks like when you set the color space and gamma in red alert in REC709 which is very saturated and bright, then your meter says something different and you chose to underexpose. Then upload into red alert and you have some areas that are generally flat and slightly underexposed that bring unwanted noise. So I tried to level up the whole ambient light, and expose a little over the areas I wanted to bring up and then bring the others down in post. I think it works. I just came from the shoot so after a good night sleep I will play with the footage and share.
PS> We have a LMD 9 inch monitor (9050 sony hdtv),a panasonic 26inch bt-lh2600, a sekonic cine L-608 zoom. I dont think its an equipment/tool problem. I think its an issue with my inexperience with this camera, plus the lack of manuals with the software, etc. And yes, my meter has batteries thank god.

[Jason Sinclair]

yea you'll work it out... let us know.

[ALLSTAR]

I shot in ASA500,T1.3,tungsten light,B16 4K 2:1 ,redcode36
http://www.reduser.net/forum/uploaded/3788_1217248737.jpg (http://www.reduser.net/forum/uploaded/3788_1217247994.jpg)
and got very noise in the shadows,This is my blue channal:
http://www.reduser.net/forum/uploaded/3788_1217249360.jpg (http://www.reduser.net/forum/uploaded/3788_1217248974.jpg)
So I need more exposure in the shadows?But in other words,RED can`t shot in high iso(over ASA500) especially under 3200K?

[David Mullen ASC]

I don't think it's reasonable to expect the blue channel to be completely noiseless in 3200K -- just as the blue layer in 500T film stock is the grainiest layer.

The question is simply does the final overall image look too noisy to you, or is it only noisy when you do things like just look at the blue channel alone.

But probably if you want less noise, you shouldn't be rating the camera at 500 ASA, though I don't think the noise at 500 ASA is that objectionable with Build 16 myself even in 3200K. But it's not noiseless.

[donatello b]

IMO, you make the final call viewing all channels at 24fps ( or 23.976, 29.97) ... i take 4k down to 2k then edit ... if i stop on a 2k still frame i do see more noise then at 24fps ... also i see less noise projected 8ft wide then on 24" lcd ( got me ?) ...

[Luke Palmer]

Not really a question about noise OR lo-light I guess so forgive me if i'm doing the wrong thing here. New to Reduser BUT increasingly being asked to shoot on the Red.
So here goes, has anyone got anything to say about shooting Day For Night on the Red. Have a commercial next week where they want to keep some detail in the sky at night in a football stadium under floodlights. I figure maybe DFN is the way to go..............

I see Fredi Callinggard in the house, with an impressively low Red no. Pick your brains?

Thanks,

Luke Palmer

[Ash Bolland]

I have to say I don;t mind that noise in the shot - I think the noise pattern in B16 is quite nice....



-
Ash

[Steve Sherrick]

It's all about what happens before the data gets recorded. When the shot is lit well, the correct aperture is set on the lens, the focus is right on, and the shutter is set properly the images can be quite noiseless and we get the exquisite 4K images we all seek.

But stray away from that process and things get interesting. It becomes very subjective. How much noise is too much? Is the noise characteristic accepted as "film grain-like". Is it just too deep in the muck?

Existing light shots are a challenge if you don't have enough stop with your lenses. As was mentioned, things can get ugly if you are having to push the image too much. I have found that quite often going beyond 500ASA is just too objectionable for me. Post can only do so much to save it. For some shots I've been okay with the noise. Others, not so much.

Expose well, post well.

[Steve Thomson]

SLR v RED

I think big big issue here is that a SLR with the same size sensor put beside a RED camera in exactly the same 'low light' will not have noise in the blacks.
If I am shooting a brightly lit model that has a background that is dropping into black, at the moment I get beautiful crisp foreground and yukky noisy background on the RED and nice solid blacks on the SLR.

[Gunleik Groven]

But in other words,RED can`t shot in high iso(over ASA500) especially under 3200K?

Yes it can and that image could have been developed significantly less noisy - if that was what you were after.

Gunleik

[Daniel Browning]

If I am shooting a brightly lit model that has a background that is dropping into black, at the moment I get beautiful crisp foreground and yukky noisy background on the RED and nice solid blacks on the SLR.

Yay, a fellow timelapse shooter!

Noise in the shadows is dominated by read noise. One component of read noise is the readout speed. At 3-10 FPS, APS-C DSLR cameras do not require the very fast readout speed of the RED ONE. Furthermore, noise is hard to compress. Therefore, if it was demonstrated that an APS-C DSLR truly has less read noise when using analog-gain ISO 1600, I would not be shocked. But I haven't yet seen anyone attempt a real comparison. (I'd do one myself if I had a RED ONE.)

There are many, many ways to perform an incorrect comparison between two similar DSLR cameras, and even more potential pitfalls when comparing with the RED ONE. Shoot RAW? Use IRIS/DCRAW (among the only converters with no noise reduction)? Use the same absolute exposure in both cameras? Used consistent artificial light (not outdoor or anything with even minute variations)? Resized both outputs to the same resolution before comparison? Used the same lens and focal point (noise in out of focus areas is more obvious to the brain)? Measured the difference in dynamic range? Resolution for a given noise level? Propensity for aliasing at that resolution? Post the RAW files from both cameras with the results?

Web sites don't get these important principles of comparison correct. Most of them compare 100% crops of default-settings raw conversions. Many people aren't aware that *every* popular RAW conversion program applies noise reduction. And the far majority of them use different noise reduction for different types of cameras. For example, D3 raw conversions desaturate in high ISO, whereas 1dsm3 is left in full color, leaving the uniformed to draw the wrong conclusion about which camera has better low light performance.

All that only covers a comparison of what the sensor is capable of. Most pepole are concerned about issues that include many other layers, such as how a given camera compares when used with certain in-camera processing (e.g. JPEG). Or when used with a certain post-processing workflow (e.g. REDCINE). But most people a more concerned with other combinations of issues.

By the way, I'm not trying to say that your comparison or findings are wrong or invalid. Only that no one has really answered the question definitively to my mind.

[Steve Thomson]

You are right that no-one has answered this question. "How do you shoot low-light on a red without getting noise"

The discussions I see regarding channels and high end cinematic techniques are sort of beside the point to me.

I know both the D3 and the RED very well.
It is very difficult to get a noisy image on the D3.
On the Red - it is very difficult NOT to get a noisy image in low light.

So to me at least - It is a very easy comparison to make accurately.

[Daniel Browning]

It is very difficult to get a noisy image on the D3.
On the Red - it is very difficult NOT to get a noisy image in low light.

First you said "SLR with the same size sensor", but now you're saying the D3. The D3 does not have the same size sensor as the RED. It has over twice the area. The laws of physics dictate that the RED ONE theoretical minimum amount of photon shot noise is more than double that of an 36x24 sensor.

As much as I give RED credit, I think breaking the laws of physics is slightly beyond their reach. ;-)

[Steve Thomson]

I shoot on both a iDmk11 and a d3 so I stand by my words.

[Esteban Sosnitsky]

Ok, it took me quite a while but I here deliver the images for you to judge and suggest. First of all I will show you a test I did in which this worst case scenario delivers a little or respectable noise over a flat blue background. It was lit with HMI from left to right delivering a gradation between the background. For me, the noise in the right part of the image is annoying, because it distracts. I know its a flat background and all but I have seen others shoot darker, or grayer backgrounds with less noise. So here is the RGB image:http://www.reduser.net/forum/uploaded/8591_1219793602.jpg

And here is the Red channel which I think shows where the annoying noise is, the blue channel looks good as I suppose HMI lighting renders better:
http://www.reduser.net/forum/uploaded/8591_1219793708.png
On the other hand, I had to shoot a greenscreen spot in which a guy jumped into a skate board. It was lit with tungsten although balance on camera set to 4000K instead of 3200K. The RGB looks like this:
http://www.reduser.net/forum/uploaded/8591_1219793809.jpg
And here is the blue channel now:
http://www.reduser.net/forum/uploaded/8591_1219793849.png
I think its very noisy. Specially in the guy's pants because they are gray almost black. Other channels look decent, but still the answer remains untold. How do you obtain less noise from these sort of images? I see a lot of noise in the blue channel of the second shot but I understand its hard if not impossible not to get that kind of noise there, BUT on the guy's pants I think its too much. Moreover the camera even started copying green from the surroundings which was not good for the chroma key. I dont know what am doing wrong, but I have seen beautiful shots made with the camera that look noise free in same sort of situations.
Any thoughts welcome.
Thanks for the patience.

[Esteban Sosnitsky]

And here is the Blue channel which I think its very clean from first shot:
http://www.reduser.net/forum/uploaded/8591_1219793658.png

[Oliver Soravia]

I`ve written some of my personal experiences with build 16 in another threat (red for the big screen), what you`re talking about is the main problem I saw with the cam while testing it extensivley for a 4K-DI printed on premiere for the big screen. So my way to avoid as much as possible that kind of noise in the darker parts of the frame, and also avoiding the clipping in the highlights, is to expose in a range of 4 stops - that`s all. It sounds very unbelievable, but my tests and every day shooting experiences for this actual feature show me the limits of MY cam... probably some other cam may produce better results, whereas I haven`t seen any real good footage without that kind of noise yet, although the CCs are mostly made very well. But this is not the way I`m looking for: to fix it in post. So I have to light for an extremly flat image and I`m pretty sure this is the only way to work with the cam to get the best results out of it. On top of that, I only use HMI-lighting and avoid any tungsten sources.

[Esteban Sosnitsky]

Hello there.
I shot another series of chromas yesterday. This ones with Build 16. Although I still feel the noise in the blue channel is still high I was gladly surprised by the improvement. I set the camera on 3200K and shot everything with tungsten. I just had no budget to do it otherwise. The chromas were excellent, one click keylight did most of the job very smoothly. We tested during the shoot, which was not very intelligent time wise but helped a lot to calm down the 3D and Post guys. I didi exactly what you have just mentioned. I shot as flat as I could. I used 3 set of silks, 2x2, 4x4, and 6x6 meters. I had about 5 stops the most of dynamic range. At first I was worried but when I opened the files in red alert and added contrast to my surprise the result was excellent. Not the way I would have done it in 35mm but at least I knew now how to use the tool. I wonder how everyone else produces those noise free images with high contrast situations. Beats me.

[Oliver Soravia]

Good question. I don`t see a chance to shoot high contrast images without getting noise - for the moment - in the future, for sure. I really had to attune to this kind of lighting and to control more than ever the parameters, but it works and the results are good. I did a small test version of the whole workflow last week: 2K-FCP editing, speedgrade xr CC in 4K, lasered on arrilaser in 4K to masterneg and printed on kodaks premiere...this is the way I will do the whole picture. The results are good. I got my deep blacks with detail in and still can see detail in the whites. No digital artifacts.

[Brian Broz]

Rather than starting a new thread I wanted to ask about "rolling off the curve in the blacks" to minimize noise.

From my experience, opening a low-light RAW file in RA/RC always has some noise (B16, ISO320) but I would imagine an unprocessed film stock would as well...but almost any recent film I've seen on Blu-ray (nowadays) has slightly crushed/glossy blacks. Just wondering how much is too much?

My question is, are most of you rolling off/crushing the blacks in RA/RC to minimize noise in the shadows as they will most defintely be in the RAW file?

I'm asking because I recently had a short turnaround low-light project (client needed the files converted by the next morning!) and all I applied was a RedSpace/REC709 look without any curves. Basically I wanted to keep the image as flat as possible but should have gotten "closer" in retrospect. Even at Prores HQ the footage was abit noisy in the shadows and looking back I should have crushed the blacks abit to minimize this.

Just looking for anyone's suggestions and experiences on "rolling off the curve" at the bottom. Basically how much are you guys doing in the first light?!

[Esteban Sosnitsky]

Oliver, your test sounds awesome. I hope we could do that kind of tests here in south america that easily. I know chile and argentina have an arrilaser but still!!! I did another series of tests today. I came to realize the only way you could get a "almost noise free" or "acceptable" image with available light and shadowed areas. Shoot in 5600K. We shot a table lit with some fluorescent light on top, and a bulb (tungsten). Set the white balance to AUTO and it went to 4098K heh! We shot it and it looked like crap. Noise was awful. Then we shot the same scene at 5600K and CC in RA to 3800K It was perfect. I think this is the way to go, and I also think this is how Gunleik and MacGregor did their tests with low light situations. I checked the channels and they were evenly exposed. So I must say I am very happy that finally I got a way around the question I originally asked in this thread which is How to avoid noise in the shadow areas. I will try and post some picts by next week of this test, you know I take some time to do that but please bare with me.

Brian, the trick is to get a little more exposure on the shadow areas. Even if it means to over expose the whole image. Why? Because in RA you can slide the exposure slide a bit down without risking your detail or noise. PLUS in the curve tool you can just roll it lower so that the noise is gone in the blacks and it looks decent. Besides B16 and ISO 320 you MUST consider shooting as near as 5600K as possible. Because of the bayer mask the sensor is much more reliable in this WB than in tungsten setting. I think I will never shoot in 3200K no more. I just cant understand why. I will test if setting the cam in 5600K and filtering it with an 80 filter in front of the lens could do the trick.
Thanks to all. peace.

[Oliver Soravia]

Estebanred. I had the same experience with the auto white function. It never corresponded with the actual color temperatur. So I started setting my color balance only by using the manual function and never lower than 5000K - whatever CT I have to deal with.

Brian. Film emulsions have a much wider latitude than digital. So the "noise" = grain (not talking about individual emulsion grain characteristics) in an emulsion starts to appear a lot later at the lower part of the curve (of a well exposed neg) than in RAW, which means, until you reach that point of "noise", you can already look back to a much wider exposer range and a much more contrasty lit image.

Best.

[Esteban Sosnitsky]

Little update on this issue.
For my workflow, I will end up in SD in 80% of the cases. So I tried and loaded the proxies in FCP transcoded with the log and transfer function with absolutely no correction whatsoever. I shot a worst case scenario shot, a glass with black background in 2K at 113fps, lots of weird noise in the black areas. But guess what, when I uploaded to FCP and previewed it in the SD CRT monitor it looked so good I cant even write about it. It was really refreshing. I wonder the thing with noise depends in: What gamma, LUT you apply to the RAW. The quality in which you monitor and where you monitor your footage. I sincerely think the computer monitor is not the best tool to monitor footage, but while 4K monitors become affordable my guess is to down convert/transcode and judge from there.
For me, tranquility came from the monitor, there was no noise there. Am very satisfied with my test results, and I dont even have my own camera yet, have been testing with friends. So, I will get my own tests as soon as the camera arrives. Hopefully sometime nextweek.

[jimhare]

I found the same thing. I was really worried about noise I was seeing but when I got down to SD it was pristine. Not surprising really, SD is pretty forgiving!

Still, I'd like to be able to minimize noise throughout the process so I'm going to experiment with some of the ideas set out here.