Jump to content

Low contrast or Dark cinematography


Bineesh Viswam

Recommended Posts

Hello friends,

Please advise,

What are the ways for noise reduction in dark or low contrast moody set up ?
Is there any camera exposure settings or lensing to achieve clear black gradations / details ?  or Do I need to light up the dark and shadow areas to avoid noise ?

Edited by Bineesh Viswam
Link to comment
Share on other sites

If the underexposed areas are too noisy, the first and most obvious approach would be to record with the lens at its widest aperture. You could also record with the lens at its widest aperture in combination with a high ISO setting, though I should note that many, including myself, would agree that bumping up the ISO is a very lazy thing to do and doesn't create a very attractive result most of the time. Now if you are not using a fast lens, or perhaps you are using a fast lens yet the shadows are still far too underexposed for the sensor to pick up cleanly, the two arguably best options would be to completely clip the shadows off in editing or add fill light to bring up the exposure of the dark areas in real time. You could even try using a low contrast filter in front of the lens. They were used religiously in the film days to pull more detail out of the shadows. The most valuable thing I've learned even with my intermediate experience, is that you can ask all the questions you want, and you can get all the answers you ask for, but you don't really fully learn it until you test the waters. You can only visualize. And since every camera sensor handles light differently, you should certainly test the waters.

  • Like 1
Link to comment
Share on other sites

  • Premium Member

I mean its all down to lighting and contrast ratio on set. The greater the ratio between your key (what you're exposed for) and the darkest areas in your shot, the more tricky it will be to clean them up and make them look good. A lot of people don't mind the background going to pure black, it's quite common these days actually. 

Nose shouldn't be much of an issue with modern cameras unless you have the ISO kicked up pretty high. So why bother kicking up the Iso when you can simply light properly to begin with? 

  • Like 1
  • Upvote 1
Link to comment
Share on other sites

  • Premium Member

We’ve had this discussion many times before. To be brief: 

1. Pick one camera ISO, either native or close to it, and stick to it.

2. Add light, both overall levels and also directional for mood. Light to a lower contrast than your intended final look.

3. Add contrast in color grading, especially in the shadows. It helps if you also preview this look on set thru a LUT, either in-camera or on a monitor.

  • Like 1
  • Upvote 3
Link to comment
Share on other sites

On 3/17/2021 at 1:57 AM, Matthew J. Walker said:

If the underexposed areas are too noisy, the first and most obvious approach would be to record with the lens at its widest aperture. You could also record with the lens at its widest aperture in combination with a high ISO setting, though I should note that many, including myself, would agree that bumping up the ISO is a very lazy thing to do and doesn't create a very attractive result most of the time. Now if you are not using a fast lens, or perhaps you are using a fast lens yet the shadows are still far too underexposed for the sensor to pick up cleanly, the two arguably best options would be to completely clip the shadows off in editing or add fill light to bring up the exposure of the dark areas in real time. You could even try using a low contrast filter in front of the lens. They were used religiously in the film days to pull more detail out of the shadows. The most valuable thing I've learned even with my intermediate experience, is that you can ask all the questions you want, and you can get all the answers you ask for, but you don't really fully learn it until you test the waters. You can only visualize. And since every camera sensor handles light differently, you should certainly test the waters.

True Indeed. Even, I dont prefer bumping up the ISO.  I would definitely try your suggestion of litting up the dark areas. Thank you for the reply Sir. ?

  • Like 1
  • Upvote 1
Link to comment
Share on other sites

On 3/19/2021 at 1:05 AM, Satsuki Murashige said:

We’ve had this discussion many times before. To be brief: 

1. Pick one camera ISO, either native or close to it, and stick to it.

2. Add light, both overall levels and also directional for mood. Light to a lower contrast than your intended final look.

3. Add contrast in color grading, especially in the shadows. It helps if you also preview this look on set thru a LUT, either in-camera or on a monitor.

Yes, Overall lighting, I would definitely try that method. Thank you for the reply Sir ?

Link to comment
Share on other sites

On 3/17/2021 at 2:20 AM, Tyler Purcell said:

I mean its all down to lighting and contrast ratio on set. The greater the ratio between your key (what you're exposed for) and the darkest areas in your shot, the more tricky it will be to clean them up and make them look good. A lot of people don't mind the background going to pure black, it's quite common these days actually. 

Nose shouldn't be much of an issue with modern cameras unless you have the ISO kicked up pretty high. So why bother kicking up the Iso when you can simply light properly to begin with? 

I agree Sir. Thank you for the reply ?

Link to comment
Share on other sites

I think its important to understand what that noise is.  The biggest goal in images is to have visually seamless gradients between pixels in representation of the scene.  One of the variables the gradients are dependent on is bit depth. The bit depth per stops of DR of the capture device are not uniform, they are actually exponential and follow the  formula 2 ^x where x is the stop of DR increment.  So for the stop between 4 and 5 stops of light, there's 2^5 - 2^4 bits that will be allocated to the gradient between those light levels per channel. So as you can imagine, lower stops get less bits and higher stops get exponential more bits. 

Noise comes from inaccuracies due to not enough bit depth to make a visually seamless gradients.  Since the camera is a machine and only can see discrete values, when it guestimates a value for a pixel each frame it can incorrectly measure the same value up or down a step simply cuz of rounding errors.  When you have so many steps that a incorrect guess is still sooo close to the next step, visually you will never see it, but if there aren't that many steps than errors are visible as noise.  So we want to increase bit depth in stops of DR that are allocated less steps in the gradient and more susceptible to noise. 

Many people never ask why we encode sensor data in a log container instead of gamma.  We want to 'linearize' the relationship some what to make the bits per stop more uniform.  When you have a function and want to undo the effect of the function you use the inverse.  Log base 2 is the inverse of 2^x, it will restore the function back to y=x.  Thats why we use a log function to reallocate the bit depth. 

When you have a 'linear' signal for the sensor data, that will not give you a linear bit depth response per stop.  It will follow the 2^x allocation, The graph shows both how few bits are allocated to the first couple stops and how fast it explodes.  You don need 200 bits+ bits to show the seamless gradation per stop, 30 steps of gradation are probably fine.    So the log transform boost significantly the signal before encoding on the low end and suppresses it in the highlights.  So variance is heavily reduced from linear sensor data.   When you change ISO down and re-expose for the scene, like the green curve, it reallocates more bit depth to the low end, increasing the fineness of the gradient, which reduces noise.  Other ways can accomplish the same thing like ETTR.  Its all the same, but you just want to place the important the detail in the scene in a portion of the log curve that captures more information per step of DR, then the gradation was be visually seamless even after transforms down the road in post.

SORRY MISLABELED THE LEGEND FOR THE GRAPH!!

ORANGE = LINEAR

BLUE = LOG

 

Capture.PNG.97041f23285bca674abd86c17ceed987.PNG

Edited by Ryan Emanuel
  • Like 1
Link to comment
Share on other sites

I am no expert, but I do not see how low bit depth is responsible for noise. My understanding is that noise comes from the randomness of electronic reading circuit, but as well from the inherent random nature of light. When light is low, its randomness becomes more obvious, even with the perfect noise-free sensor and 1-zillion bit AD converter. Hence the best solution is to add more light.

I would think that low bit depth could even reduce noise: with levels more distant from one another, there is less digital levels to choose from. Analogue levels slightly varying due to noise will lead to the same rough digital level. But this could create banding. In the audio domain, "dithering" is the art of adding noise to reduce low bit depth artifacts.

Edited by Nicolas POISSON
  • Like 1
  • Upvote 1
Link to comment
Share on other sites

  • Premium Member
On 3/18/2021 at 3:35 PM, Satsuki Murashige said:

We’ve had this discussion many times before. To be brief: 

1. Pick one camera ISO, either native or close to it, and stick to it.

2. Add light, both overall levels and also directional for mood. Light to a lower contrast than your intended final look.

3. Add contrast in color grading, especially in the shadows. It helps if you also preview this look on set thru a LUT, either in-camera or on a monitor.

Agreed. But I'll expand on point 1 about selecting ISO.

For dark scenes with lots of shadow, set your ISO at a stop under the native ISO (or more if you prefer). This sounds counter-intuitive as others suggest raising it. But shooting at 400 when your camera is 800 native, this forces you to add more light to the sensor and overexpose the log footage. The result is cleaner shadows in post. I do this all the time on Sony cameras with dark scenes. Their cine EI mode makes it easy.

Why overexpose by a stop? Because if you're in a dark scene, then the highlight latitude isn't being utilized anyway.

  • Upvote 1
Link to comment
Share on other sites

Quote

Why overexpose by a stop? Because if you're in a dark scene, then the highlight latitude isn't being utilized anyway.

I've done this for dark, moody situations a lot of times. Even if I want them to look really dark, I've overexposed by two stops, so that in case they want to raise them later in post, there's even more latitude to do that. It has worked for me with RAW recordings with Alexas and Reds.

  • Like 1
Link to comment
Share on other sites

  • Premium Member

The simplest thing is to pick a lower ISO if you want less noise and then light it the way you want.

Next simplest thing is to the ISO you want (if higher) but create a LUT with more contrast in the shadows so that you'll be forced to add more fill light to get the balance you want.  But I think the first option is better.

Noise comes from underexposure. Shadows are naturally less exposed than highlights.  But the technology shouldn't be driving the creative lighting choices.

  • Like 1
  • Upvote 1
Link to comment
Share on other sites

  • 3 weeks later...
On 3/26/2021 at 11:49 PM, Ryan Emanuel said:

I think its important to understand what that noise is.  The biggest goal in images is to have visually seamless gradients between pixels in representation of the scene.  One of the variables the gradients are dependent on is bit depth. The bit depth per stops of DR of the capture device are not uniform, they are actually exponential and follow the  formula 2 ^x where x is the stop of DR increment.  So for the stop between 4 and 5 stops of light, there's 2^5 - 2^4 bits that will be allocated to the gradient between those light levels per channel. So as you can imagine, lower stops get less bits and higher stops get exponential more bits. 

Noise comes from inaccuracies due to not enough bit depth to make a visually seamless gradients.  Since the camera is a machine and only can see discrete values, when it guestimates a value for a pixel each frame it can incorrectly measure the same value up or down a step simply cuz of rounding errors.  When you have so many steps that a incorrect guess is still sooo close to the next step, visually you will never see it, but if there aren't that many steps than errors are visible as noise.  So we want to increase bit depth in stops of DR that are allocated less steps in the gradient and more susceptible to noise. 

Many people never ask why we encode sensor data in a log container instead of gamma.  We want to 'linearize' the relationship some what to make the bits per stop more uniform.  When you have a function and want to undo the effect of the function you use the inverse.  Log base 2 is the inverse of 2^x, it will restore the function back to y=x.  Thats why we use a log function to reallocate the bit depth. 

When you have a 'linear' signal for the sensor data, that will not give you a linear bit depth response per stop.  It will follow the 2^x allocation, The graph shows both how few bits are allocated to the first couple stops and how fast it explodes.  You don need 200 bits+ bits to show the seamless gradation per stop, 30 steps of gradation are probably fine.    So the log transform boost significantly the signal before encoding on the low end and suppresses it in the highlights.  So variance is heavily reduced from linear sensor data.   When you change ISO down and re-expose for the scene, like the green curve, it reallocates more bit depth to the low end, increasing the fineness of the gradient, which reduces noise.  Other ways can accomplish the same thing like ETTR.  Its all the same, but you just want to place the important the detail in the scene in a portion of the log curve that captures more information per step of DR, then the gradation was be visually seamless even after transforms down the road in post.

SORRY MISLABELED THE LEGEND FOR THE GRAPH!!

ORANGE = LINEAR

BLUE = LOG

 

Capture.PNG.97041f23285bca674abd86c17ceed987.PNG

Wow... its a detailed technical information. thanks for the effort Ryan. Exteremely delighted.

Link to comment
Share on other sites

  • 1 month later...
On 3/28/2021 at 7:52 PM, David Mullen ASC said:

The simplest thing is to pick a lower ISO if you want less noise and then light it the way you want.

Next simplest thing is to the ISO you want (if higher) but create a LUT with more contrast in the shadows so that you'll be forced to add more fill light to get the balance you want.  But I think the first option is better.

Noise comes from underexposure. Shadows are naturally less exposed than highlights.  But the technology shouldn't be driving the creative lighting choices.

Thank you so much for the clarification, Sir. Honoured!

Link to comment
Share on other sites

  • 5 weeks later...
On 3/27/2021 at 9:58 AM, Stephen Sanchez said:



Why overexpose by a stop? Because if you're in a dark scene, then the highlight latitude isn't being utilized anyway.

Sorry for the ignorance, but trying to wrap my brain around this still. Do you mean overexpose by a stop based on your lower ISO (400 for example) or for your 800 ISO? So overexpose the 400 so it's essentially the same as your 800, but with cleaner shadows because of the lower ISO?

So say you're shooting a scene with a single light, and want clean shadows. The light reads a T/8 at 800, so a T/4 at 400, so you would want to be at a T/2? Have I got that right?

Link to comment
Share on other sites

  • Premium Member

Don't apologize. It's complex at first glance.

So on video cameras that record log, the log image is the full range of the sensor. It records this file at the set ISO it was manufactured at (base ISO). That ISO can't actually change.

When you view the image with a LUT (which you need to do), the LUT can only see a range of that log image. So when you push your ISO, you're viewing the shadow section of the log image, which is why its noisier. And this leaves the mid and highs of the camera's range unused. You need more light on the sensor to utilize the dynamic range its capable of, which will inherently give more detail in the shadow areas. So when you view the log in a lower ISO LUT, and light it that way, you are feeding the sensor more light. Switching between ISOs doesn't do anything to the log image, its for you to expose along the range of the sensor.

So, for a cleaner image, shoot scenes with lots of shadow or darkness at a lower ISO (by 1 or 2/3 a stop). And for scenes with lots of white or hot areas, use the base ISO.

If you are recording Rec709, the same principle applies because the camera is printing a LUTed image from log in realtime.

This is different for stills cameras and I don't understand why. Stills cameras have a base ISO (say 800), but you can shoot at 100 and have a cleaner image and the most dynamic range.

Link to comment
Share on other sites

Thanks for the explanation, Stephen. So the lower ISO is more for me to "overexpose" the image and thereby utilize more of the higher end of the dynamic range rather than actually changing exposure with it? I would think that the cleaner image would come from the lower ISO, and have less noise in the shadows because of that, but if I'm shooting in LOG, changing the ISO won't matter? So then I have to overexpose by a stop or so in order to get a cleaner image? I wasn't aware that changing the ISO didn't matter in LOG, thanks for sharing your knowledge!

Link to comment
Share on other sites

  • Premium Member

If you shoot log on a camera like the ARRI Alexa, ISO is baked in to the recording (usually ProRes) — it’s only if you shoot Arriraw that ISO is just metadata. However, either way, your ISO setting affects the light levels and exposure you use whether or not the image processing to create that ISO setting is done in-camera and recorded as log — or done in-camera only for display but not recorded in raw. Either way, you give the sensor twice as much light by rating the camera at ISO 400 instead of ISO 800.

Link to comment
Share on other sites

On 7/8/2021 at 1:22 PM, Mateusz Czopek said:

Thanks David and all for your help. Still wrapping my head around it, but I appreciate all the wisdom!

Thinking about it this way may help your head conceptualize it. The point of rating a cine cam lower in iso is to be able to expose low key stuff to your desired look on your monitor the day of on set, and to make post’s life easier by giving them a file that resembles your intention. Whether you shoot at 800 or 400 iso, you want to create an image in your monitor that resembles your finished goal.
 

Say that your shadows, skin tones, etc are all where you want them at t/4 @iso 800. If you lower the iso to 400, you now need to be at t2.8 to get your image to match the iso 800 one. They are the same look, but since you need double the light for iso 400, you get a stop better noise. Iso 200 would give you two stop cleaner shadows and need a t2 to look the same. Your highlights will clip the same amount of stops sooner, but usually that isn’t as important for those types of underexposed scenes. 
 

you can do the same thing by creating a lut that is underexposed by a stop, and keeping the camera at iso 800. You’ll still need to open your aperture or add twice as much light to make the image match the standard lut. The downside of doing this method is it can leave interpretation up to the color grader, even if you send the lut as a reference. It’s also more work than doing it in camera IMO, and when shooting pro res there may be a different dynamic range response as opposed to a raw file that can be edited in metadata. Otherwise you would get the same situation. Cleaner shadows, but highlights clip faster. 

Link to comment
Share on other sites

  • Premium Member
On 7/10/2021 at 12:31 AM, Adam Allington said:

Say that your shadows, skin tones, etc are all where you want them at t/4 @iso 800. If you lower the iso to 400, you now need to be at t2.8 to get your image to match the iso 800 one. They are the same look, but since you need double the light for iso 400, you get a stop better noise. Iso 200 would give you two stop cleaner shadows and need a t2 to look the same. Your highlights will clip the same amount of stops sooner, but usually that isn’t as important for those types of underexposed scenes. 

Perfect explanation! Kudos Adam.

@Mateusz Czopek, so with the above scenario, shooting your dark scene at 400 instead of 800 makes it darker for you. Your response is to either give the scene more light with the iris, or hotter fixtures on set.

The more you shoot, the more you'll understand what you can get away with, and what you shouldn't. And each camera make varies a little with the noise levels. So get to know your tool.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...