FIGHTING AGAINST NOISE or how to get all out of RED

[Kim Frank]

Im a director and DOP for Musicvideos from Germany.
Although this is my second post, I’ve benn around a lot in this Forum and must thank all of you for sharing your knowledge.
I shot six music Videos and a Shortfilm on Red within the last six months.
Because all of those Projects were really low budget and I dont own a RED, I never had time to do Tests and had to learn while Shooting and Postproduction.
One thing that came into my way was this ugly digital noise, happening when underexposing or shooting under tungsten light (or both) and I just would like to clarify if I finally do get it right. So here is what I think I know and I would love you all to correct me if Im wrong:

Many people around me (even shooting faeture Films with RED) think that it doesnt matter how you white balance or set your ISO cause its all metadata. They are right in terms of that you can change the settings later in Alert or Cine. But what they dont keep in mind is that the RED Sensor has its own native settings with 320 ASA and balanced to 5000 kelvin and is recording in RGB. If you now shoot under tungsten Light without filtering you do underexpose the blue channel witch means it isnt recorded with the full Bit rate. If you want to balance the footage back to 5600K in post or add a lot of contrast you have to bring up the low resolouted blue channel wich brings up the noise.
To avoid this while shooting under tungsten (or other low and really high tempretures) you have to gel the lights or put a filter in front of the lense to get it at leat balanced close to 5000K. Even if you want the picture to have a warm (or greenish or blueish) look, you should try to record all of the three channels balanced equaly and tweak the desired look in post using all the possible information of each Channel.
To get rid of noise caused by underexposure that needs to be brightend later on you should expose to the right (ETTR) always pushing the Histogram right before clipping in the Highlights or letting just the desired hot spots clip. And you should keep the dynamik range of the RED in mind and use more fill light in critical situations. Even when shooting dark night scenarios its better to not underexpose the full way. Just watch out nothing clips and do the desired darkening in post.

Hope Im right so far and didnt miss anything.
Now my questions I didnt have the chance to find out by my own til now:

1. Does the colour of the subject change the exposure of a channel.
For example if someone stands in front of a red wall or against blue sky.
Or is the exposure of the channels just about the tempreture of light.

2. What about mixed lighting sources or if the color should change.
Pretend you want to shoot someone in his car standing in front of a traffic light by night and you want to see the red light on his face, than let it change to green and as he drives along youre getting the normal Streetlights again.
I think you would have to use a filter to get the streetlamps right the all over leght situation.
Now you would place one red and one green geled fixture in front of the car and switch them on and off to enhance the effect.
But what happens to my RGB channels by doing that?
Wouldn’t while the red light is switched on the green and blue channel get massivly underexposed? Doesnt it matter in terms of noise, because I dont want to bring everything back to „normal“ white? And what about the contrast of the scene if the blue channel is suddenly underexposed, would that cause noise?
Or for another example what to do in a colourful lighted Club or at a Concert?

3. I don’t really know if this belongs here, but I think its importand in terms of noise:
Cutting in Final Cut but grading in After Effects I render out full res TIFFS of the edit.
What is the best Luma and gamma space when converting the r3d files?
Having both set to Redspace adds a realy contrasty curve for my opinion so sometimes highlight information gets lost. Setting gamma to linear light, i suppose doesnt add any curve but the picture gets a lot darker as well. Now you can increase exposure what does infect the colours. Or you bring up the Brightness what I do right before the Histogram clips. Is it right to do these adjustments in ALERT? I tested some pictures of a talent against bright sky where his face was very underexposed and I wanted to get it to normal. I rendered out in linear light without Adjusments and did all the brightnig in After Effects. My feeling was that this way caused more noise than doing it before rendering in Alert.
Is that because the brightness slider in Alert really uses all the RAW information possible?
And does it make sense to expand the Histogram by pushing the blacks back down to the low end using Curves in Alert?

I have to appologize for this long post, but I had a lot to think of in the last months.
Looking forward to read your experiences and suggestions.
Regards
Kim Frank

 

[Vigen Vartanov]

Answer 3.
Best gamma for render out is : REDLOG .
and better to render out footage in DPX files.

 

[Stepan Kment]

Answer 3.
Best gamma for render out is : REDLOG .
and better to render out footage in DPX files.

DPX as a final output is certainly OK. I add my thougts on using it as a format for transfer among various software, editing>color grading or vfx etc:

As an input for color grading and exposure changes, there still is imho a problem: RED has 12-bit CMOS A/D sampling I believe. Conversion to most common 10-bit DPX, of any kind, no matter log or lin, causes lowest important bits to be lost. DPX is good, but still not optimal for this. Even worse, it must be carefully handled: common log685 format actually has large headroom for near to black and over the white colors, so typical 12-bit interval 0..4095 may be compressed into something like 90..685 only, like 9+bits...
That's why I'm thinking of Cineform 12-bit 444 mode - OK, it's compressed, but has all the tools to convert from r3d and to dpx.

Loosing precision may cause banding in the picture, if exposure is changed significantly. Be careful not to decimate bits if it is not really necessary prior .. at least first light exposure and color ballance correction, otherwise you can run in trouble.
---
side note on color and the CMOS photosites - think of each photosite as an independent measurement device. That is, R, G and B photosites generate equal output values, if they are lit with 5000K mix of light frequencies. You correctly mention 3200K lighting causes blue channel to be underexposed, bad is silicon is not very sensitive to these close to ultra-violet part of spectrum light frequencies (is sensitive to infra-light on the other hand, often bad, too, that's why we use infra-shielding as a filter as it may deteriorate the picture, too) and so noise is quickly introduced, the reason why profi monitors have an option to view just blue channel itself, it's typically the easiest way to judge amount of a noise in the picture.

 

[Stepan Kment]

1. Does the colour of the subject change the exposure of a channel.
For example if someone stands in front of a red wall or against blue sky.
Or is the exposure of the channels just about the tempreture of light.

as I wrote above. Think of single each cell. The channel is simply a collection of the cells of given color. And histogram is built from all these cells. A cell obviously is designed to accept both shadow and highlight.

The noise means - if you point the chip at absolutely flat and evenly lit wall, you won't get identical reading from each cell. And noise is most visible in shadows on large underexposed areas.

That said, blue person in front of a red wall, all lit with 5000K light, 3200K light or any other - will generate underexposed green channel, obviously - since there is no green light bounced off to the camera.

 

[Florian Stadler]

You can certainly shoot perfectly fine under tungsten light. You must know where the noise floor is (~3.5 stops under key) and not put any information that's not pure black under that noise floor. You will see information in your monitor but it is not useable information, you have to crush it or live with noise or noise reduction software.

 

[M Most]

As an input for color grading and exposure changes, there still is imho a problem: RED has 12-bit CMOS A/D sampling I believe. Conversion to most common 10-bit DPX, of any kind, no matter log or lin, causes lowest important bits to be lost. DPX is good, but still not optimal for this. Even worse, it must be carefully handled: common log685 format actually has large headroom for near to black and over the white colors, so typical 12-bit interval 0..4095 may be compressed into something like 90..685 only, like 9+bits...

That is not the case. Redlog is specifically designed to properly map the 12 bit linear data into a 10 bit log curve. The lower 8 stops of information is retained with full precision, as this is where the vast majority of the visual information is contained. The upper 4 stops - the brightest areas of the original image - are mapped with less precision, but there is no significant loss of detail as this area usually represents "super white." Graeme can explain this in detail, but suffice it to say that a proper transcode to DPX from a R3d file using Redlog yields a very accurate representation without any significant information loss.

 

[Antoine Fabi]

That is not the case. Redlog is specifically designed to properly map the 12 bit linear data into a 10 bit log curve. The lower 8 stops of information is retained with full precision, as this is where the vast majority of the visual information is contained. The upper 4 stops - the brightest areas of the original image - are mapped with less precision, but there is no significant loss of detail as this area usually represents "super white." Graeme can explain this in detail, but suffice it to say that a proper transcode to DPX from a R3d file using Redlog yields a very accurate representation without any significant information loss.

Makes sense, but do you "burn" the 10bit DPX without applying any gamma curve ? ...or do you care to carry most info in the middle of the histogram before burning DPX 10 bit ?

Antoine

 

[Stephen Pruitt]

Okay, okay. . . hang on. . .

Florian. . . are you saying that to avoid noise, I need to keep my dynamic range down to 3.5 stops???? Surely that's not what you meant. . . but that is what I read.

Stephen

 

[Birns and Sawyer]

I think he's saying that the noise floor is ~3.5 stops below key, and you'd have a few stops over key to work with.

In my experience, an ISO of 160 and having sufficient light to rate as such keeps the blacks nice and clean without crushing them any more in post. If you're going to crush a bit, then you could do ISO 200 or 250.

 

[M Most]

Makes sense, but do you "burn" the 10bit DPX without applying any gamma curve ? ...or do you care to carry most info in the middle of the histogram before burning DPX 10 bit ?

I don't understand the statement. Redlog IS a gamma curve. It is designed to match the characteristics of the sensor in terms of how it codes its output values. That's the advantage of having the manufacturer design the curve, much as it is with the Genesis and Panalog. When you use Redlog, you don't do any color correction prior to the transcode (other than protecting any potentially clipped highlights), you do it after the conversion.

 

[Antoine Fabi]

"...(other than protecting any clipped highlights)..."

That was exactly my question...

I should have said "applying any adjustment" instead of "applying gamma curve".

Thanks.

Antoine

 

[Florian Stadler]

Okay, okay. . . hang on. . .

Florian. . . are you saying that to avoid noise, I need to keep my dynamic range down to 3.5 stops???? Surely that's not what you meant. . . but that is what I read.

Stephen

No, what I meant is that from a spotmeter reading of 18% grey (key) you have about 3.5 stops of underexposure until you reach black without significant noise.

 

[ScottyC]

Hello,
I know I am only listed as a Junior member and haven't bought a RED camera yet, but I have a good bit of experience with digital cameras and thought I would help you down the same path I took when I started using digital cameras. So to give you some insight that might help you in your pursuit to reduce the noise, I want to provide a link that explains in details all the quirks of digital photography.
http://www.cambridgeincolour.com/tutorials.htm

The CMOS sensor in the RED cameras is similar to the CMOS sensors in professional digital cameras like Nikon and Canon. They are Bayer pattern sensors that read RGB, but have 2 green photosite sensors for every 1 red and 1 blue (something about our eyes being more sensitive to green light). This means that an image with more green in it is more likely to be cleaner than one with very little green. Noise is created in the image when the photons of light in the overexposed areas escape the photosite cavity and bounce into the photosite cavities of the underexposed areas. This only affects large underexposed areas because the areas that are exposed correctly receive enough light from the image source to correctly record the color. If you record footage with a cap on the lens, you will still see noise because there will always be a few photons of light bouncing around on the sensor. To reduce noise, a lower ISO (like 100 instead of 320) would make the sensor less sensitive to light and those stray photons would not read in the image. But this also means that your subject/image must have more light for your sensor to read or it will look terribly underexposed (so shooting at night becomes very difficult). Another thing to consider is a lower f/ stop like f/1.2 instead of f/3.5 because this would also provide more light to the sensor and reduce the noise, but the downside is that your focal range becomes more limited and moving subjects become much more difficult to keep in focus and those lenses are generally much more expensive (but I assume you know this already).

Also, the noise in your image becomes much more apparent when you try to make major corrections in post. Like trying to brighten and saturate your image in post will make the noise stand out much more, regardless of whether you are working in RAW or another format because the software is still utilizing the same sensor data. Also, I noticed someone mentioned the use of noise reduction software. Many people use this kind of software, but it's good to know exactly how it reduces the noise. It generally does this two ways: 1)by quickly scanning and analyzing the image for color and trying to even out that color (like a red circle with a tiny green dot, the green dot would be turned red) and 2)by adjusting the sharpness of the image to a softer look. So the noise reduction software might not be a great option if you are looking for something specific (like a really sharp image of a nighttime city skyline). To avoid all this, it's just best to try and expose or overexpose the image. The more light the sensor gets, the better the image quality with less chance of noise. If your used to the film medium, this might be a difficult switch because you are used to exposing for shadows and the digital medium should be exposed for highlights because we want the sensor to get more light.

Someone mentioned the different options for rendering out your r3d footage and the differences in bit-depth. Anytime you render footage it is usually best to use the highest quality (but you will spend lots of time rendering the footage). 12 bit footage has more available colors than say 8 bit (i.e. 8 bit = 256 colors, 16 bit = 65,536 colors)... so your images will have better color gradients with a higher bit depth. The human eye can only see about 10 million colors, so anything over 24 bits is excessive. Anytime you compress or render your footage into a different format, you are bound to lose some data... it's just how much you are willing to lose and whether that loss will affect the quality of your work. 10-bit DPX is a very high quality format and should be sufficient for any high-quality product, whether it is going to be broadcast, distributed on film, or even just put on DVD. If you are just shooting for Web, DPX is overkill. I remember one member in here who was just using the low-res quicktime files for their web project and it rendered out just fine for their needs. Cineform is also a very nice compression format. I believe it is a floating-bit compression, so you still maintain high-quality footage (meaning it allows the file size of the footage to be larger if there is more image data where-as other encoders will have a fixed compression like 25Mbs). Those encoders work fine for cameras from Sony and Panasonic because they already compress the footage down to sizes like 25Mbs while shooting where-as RED compresses to a much larger size allowing more image data to be recorded.

You also mentioned using different colors in your shoot to convey the traffic light changing color on the subject in a car... have you considered creating this color effect in your post-production grading? By doing that, you can properly expose all RGB channels and avoid noise and sensor issues and still get the color look you want from your final footage. You would just have to play with the color grading a bit to achieve the effect without making the image look weird. This will add extra time to your post-production work, but your image would have less noise and look cleaner.

I am sorry my response is long, but I hope this info was helpful and not redundant (I do not know how much experience you have with digital sensors, so some of this might seem like amatuer knowledge).

 

[Philip Powell]

The changing of white balance is not always a neutral adjustment. This is easy to demonstrate, take a clip (shot with tungsten, and WB of 3200 set) into Red Alert.

Just look at the blue channel, and shift the WB around, setting it to 5000, and it makes it less grainy, (native) and shifting it lower gains the blue channel and increases noise. Shift it to 1700) or similar and you'll really see some noise.

In daylight balanced and illuminated shots I have not found this to be so much of an issue.

 

[Kim Frank]

Thank you all for your Time and response

@Vigen: Doesnt a DPX file have less bits than a TIFF?

@ Stepan: So just the reflecting light is what is messured by the Histogramm?

@ Florian: I do understand the meanig of NOISE FLOOR
but if you need to balance tungsten light situations to normal grey you need to lift up the underexposed Blue channel wich increases noise.
For me noise reduction software doesnt work cause it just changes the colour of pixels or softens the image.

@mmost: I thought the sensor would record linear - more details in the highlights -
why should I compress them turning into DPX?

@ Redline: Thanks for the link... I thought I would know digital Photography, even the RED till I read more of these threads and got those noise Issues.
Doing the Traffic light effect in post would effect the whole image, what I dont want.

I Still dont get the RGB balance thing and also dont get this confusing ISO/ASA discussion.
I thought the Sensor would be 320...
Then they say Increase the ISO in High Contrast Scenes
Set it down in low dynamic range...
Wouldnt the real captured RAW DATA always stay the same instead of
me opening the lense when setting ISO down (trying to get more shadow information they say) or shutting it down if Im increaing the ISO (saving Highlights - they say)
Excuse me being confused.
What actually does the ISO(ASA Setting in Camera change - if its Metadata and the native reading is 320 it shouldnt matter, but the Raw Histogramm and Meter should be exposed to the right...

 

[ScottyC]

Redline: Thanks for the link... I thought I would know digital Photography, even the RED till I read more of these threads and got those noise Issues.
Doing the Traffic light effect in post would effect the whole image, what I dont want.

I Still dont get the RGB balance thing and also dont get this confusing ISO/ASA discussion.
I thought the Sensor would be 320...
Then they say Increase the ISO in High Contrast Scenes
Set it down in low dynamic range...
Wouldnt the real captured RAW DATA always stay the same instead of
me opening the lense when setting ISO down (trying to get more shadow information they say) or shutting it down if Im increaing the ISO (saving Highlights - they say)
Excuse me being confused.
What actually does the ISO(ASA Setting in Camera change - if its Metadata and the native reading is 320 it shouldnt matter, but the Raw Histogramm and Meter should be exposed to the right...

Sorry my suggestion won't work for your project... it was just a thought.

I know my link is for digital photography and not digital cinematography, but the information provided is pretty universal. It's a good place to start to get some answers about digital acquisition.

Someone please correct me if I am wrong but I believe that Jim and his team were trying to make the transition to a digital format easier to the DP's who are used to selecting the different films for their needs, but this might have created some confusion. Just like films sensitivity to light is rated by ASA, so is the sensor but by ISO (even though the RED says ASA) and noise is the equivalent to film grain except noise looks bad. I believe RED's ASA 320 is actually ISO 320, but I could definitely be wrong... the RED team would know the true ASA/ISO equivalencies in their camera. I think there are several different post in here about the ASA/ISO, but I don't have the links. You mentioned 320 being the sensors "native" reading, but digital sensors don't really have a "native" ISO... but they do have an ISO range. Some sensors can range in ISO from 50-400 and others from 100-25,600, but it all depends on the quality of the sensor and the hardware/software supporting it. But the lower ISO (like 100) produces less noise. 320 might be the setting to achieve the 66db (signal-to-noise?) dynamic range, so increasing the sensors sensitivity to a higher ISO (or ASA) ,like 800, would affect that range and show more noise. This is why the group told you about adjusting the ISO for contrast because you want the better range of the sensor for those shots.

The RAW footage is not quite what you think. If you adjust the sensors sensitivity on the camera, you cannot change that in post (much like film). Same goes for the f/stop of the lens and the focus of the lens. You can adjust the footage to make it brighter or darker than the original, but the noise in the footage becomes much more apparent when you try to make more drastic compensation from the original footage. Darkening an image is usually easier with digital than brightening because noise hides in the underexposed areas, but you can have blown-out highlights in your image if you overexpose. High contrast scenes are always the hardest and you would have to choose what you would rather see, a silhouette or a blown-out background. RAW footage just means that the camera is just recording what it sees instead of compressing information into a final format to save space. Basically, you have the ability to make better corrections to the footage with the RAW data than you can with compressed data because you have all the image information available to adjust instead of a reduced amount of information from a compressed format. But ultimately the RAW data will NOT be the same if you make certain changes on the camera while shooting. RAW data is great for making changes to the temperature, contrast, and other adjustments... but there are some changes that you cannot make to the RAW data.

Don't worry about being confused, even I am still learning new things and thats what these forums are for. Hopefully I answered more questions than I created.

 

[Esmaeil Neisi]

thank's to all, nice info

 

[Steve Sanacore]

Sorry my suggestion won't work for your project... it was just a thought.

I know my link is for digital photography and not digital cinematography, but the information provided is pretty universal. It's a good place to start to get some answers about digital acquisition.

Someone please correct me if I am wrong but I believe that Jim and his team were trying to make the transition to a digital format easier to the DP's who are used to selecting the different films for their needs, but this might have created some confusion. Just like films sensitivity to light is rated by ASA, so is the sensor but by ISO (even though the RED says ASA) and noise is the equivalent to film grain except noise looks bad. I believe RED's ASA 320 is actually ISO 320, but I could definitely be wrong... the RED team would know the true ASA/ISO equivalencies in their camera. I think there are several different post in here about the ASA/ISO, but I don't have the links. You mentioned 320 being the sensors "native" reading, but digital sensors don't really have a "native" ISO... but they do have an ISO range. Some sensors can range in ISO from 50-400 and others from 100-25,600, but it all depends on the quality of the sensor and the hardware/software supporting it. But the lower ISO (like 100) produces less noise. 320 might be the setting to achieve the 66db (signal-to-noise?) dynamic range, so increasing the sensors sensitivity to a higher ISO (or ASA) ,like 800, would affect that range and show more noise. This is why the group told you about adjusting the ISO for contrast because you want the better range of the sensor for those shots.

The RAW footage is not quite what you think. If you adjust the sensors sensitivity on the camera, you cannot change that in post (much like film). Same goes for the f/stop of the lens and the focus of the lens. You can adjust the footage to make it brighter or darker than the original, but the noise in the footage becomes much more apparent when you try to make more drastic compensation from the original footage. Darkening an image is usually easier with digital than brightening because noise hides in the underexposed areas, but you can have blown-out highlights in your image if you overexpose. High contrast scenes are always the hardest and you would have to choose what you would rather see, a silhouette or a blown-out background. RAW footage just means that the camera is just recording what it sees instead of compressing information into a final format to save space. Basically, you have the ability to make better corrections to the footage with the RAW data than you can with compressed data because you have all the image information available to adjust instead of a reduced amount of information from a compressed format. But ultimately the RAW data will NOT be the same if you make certain changes on the camera while shooting. RAW data is great for making changes to the temperature, contrast, and other adjustments... but there are some changes that you cannot make to the RAW data.

Don't worry about being confused, even I am still learning new things and thats what these forums are for. Hopefully I answered more questions than I created.

I think there are a few errors here in your explanations. It would be best if Stuart or another RED expert would clear them up but the two errors I think I see are:

1.The RED raw format is a compressed format. The uncompressed data would be prohibitive to try and save to a CF or small hard drive in real time.

2. Any adjustments to the image while shooting, ISO, color balance etc. Only effect the proxy files and does not 'bake' anything into the raw files. I'm not sure you can set anything on the camera other than frame rate, shutter speed and data rate, that will change the raw file instead of just metadata.

Let me know if i am mistaken. Thanks.

 

[ScottyC]

You are correct Steve, The RAW data is compressed... but the compression is much less than other compression types. I have always been told that RAW footage contains more image information that is usable in creating a final "look" to you footage than with something like the Sony or Panasonic cameras produce (because they are limiting the file size to a much smaller amount with less bit depth and creating a pre-composed "look" from the camera settings). I apologize if I conveyed the incorrect information.

I would still argue that the ISO affects the image and not just the proxy because the sensor sensitivity will affect the final image. You can put this theory to the test by shooting the same scene twice, once with the ISO at it's highest (say 800) and then at it's lowest (say 100) capability, but adjust the shutter, etc to have a proper exposure. When you take the footage and adjust it to make it much more saturated... the noise from the higher ISO footage will stand out considerably more. I say this because the RAW footage is supposed to capture just what the sensor sees and not any in-camera alterations or "looks"... so if the sensors sensitivity produces more noise it would not be something that you can fix in post - meaning it's not just a proxy. Would you all agree?

 

[Graeme Nattress]

ISO is purely metadata - it does not effect the sensor.

Graeme