ivan friedman
Well-known member
- Joined
- Apr 18, 2012
- Messages
- 331
- Reaction score
- 0
- Points
- 16
- Location
- Lausanne, Switzerland - Roma, Italia
---
Last edited:
Welcome to our community
Be a part of something great, join today!
Dragon: resolution vs Dynamic range.
- Thread starter ivan friedman
- Start date
- Banned
- #2
Robert Ruffo New
Banned
Yes and no - the smaller the pixel vs display (a single 6k pixel is tiny when converted to 1080p, much less than 1/5th of one pixel) the smaller the noise, and the more that noise will tend to be averaged out into something smooth.
Imagine now a 480p SD capture. Even a tiny amount of subtle noise would be a bunch of huge off-color dots and be extremely visible.
Thats' why it's not pixel size that matters, it's pixel coverage as a percent of sensor surface. Pixel size only means less noise in simple way if you are looking at a single pixel and its accuracy, not the overall accuracy of the pictures all pixels combined produce.
Not to mention, the more pixels you throw at a noise reduction algorithm such as Neat video or Resolve's new motion-based tool, the more nosie can be reduced without killing sharpness or introducing a plasticky look.
Imagine now a 480p SD capture. Even a tiny amount of subtle noise would be a bunch of huge off-color dots and be extremely visible.
Thats' why it's not pixel size that matters, it's pixel coverage as a percent of sensor surface. Pixel size only means less noise in simple way if you are looking at a single pixel and its accuracy, not the overall accuracy of the pictures all pixels combined produce.
Not to mention, the more pixels you throw at a noise reduction algorithm such as Neat video or Resolve's new motion-based tool, the more nosie can be reduced without killing sharpness or introducing a plasticky look.
Neil Cronin
Well-known member
I am not a sensor engineer, but this is an interesting topic for sure. I'd appreciate it if folks could correct any mistakes in my understanding:
The size of a photosite is one of many factors in its sensitivity. Sensitivity dictates the minimum amount of light required to produce information to be recorded.
Dynamic range is the difference between pure black and pure white values that a sensor can read, as well as the number of interim gradations that produce perceptually differential information.
Noise floor is a perceptual thing -- there is some amount of noise in any real-world sensor output. But there is general wisdom about what humans can perceive.
My understanding is that scaling up a sensor's photosite size would not necessarily increase dynamic range, it would only reduce the minimum amount of light required to register a signal, improving its sensitivity. It would also reduce the clip point of the sensor, though, by whatever factor you scaled up the size of the photosites. So if you made the Dragon more sensitive, you would just be clipping skies at a less-bright level, which is not necessarily desirable.
Increasing sensitivity while artificially reducing the output gamma curve in the typically-noisy-region would reduce the perceptual noise floor, since the sensor would be throwing away information that it "knew" it could not reliably capture without unwanted noise. This would also impact the perceived dynamic range, since a smaller amount of information would be recorded at the dark end of the continuum, since it was thrown away because it was probably way too noisy.
It's also my understanding that although these things are, in an academic way, mostly unrelated, in the real world there is a ton of practical overlap and complications, and the real magic is in tweaking these things to get good pragmatic results. This is one of the jobs of the gamma pre-sets that control the debayer process, among many others.
From what I've seen of Dragon footage, it does look like sensitivity is increased, and it also looks like dynamic range is increased. But these things are mostly unrelated. It's hard to tell about the noise floor without more information about what was happening at the time of capture, or by seeing side-by-side footage with another camera that you are familiar with. But you could theoretically increase sensitivity while also making the relative noise floor worse, and vice versa. You could also increase dynamic range, but make sensitivity worse.
MX does not have great sensitivity compared to other digital camera sensors I've used. However, it has great dynamic range compared to other sensors I've used.
I think there was a chart posted that showed Dragon getting 19 stops of theoretical dynamic range, or something. But that doesn't help if, in reality, the bottom 2-3 stops are below the noise floor. It also doesn't help if the darkest stop that the sensor is able to record is as bright as the sun, because then you could never use the camera on planet earth to film stuff. To fix the noise floor, you might have to crush out the bottom 2-3 stops in software, either in firmware on the camera or in the debayer defaults. To fix the minimum sensitivity, you might have to crank the gain from each photosite, which would then make the overall noise floor worse. It's all a bit like a giant 20-sided rubik's cube, except there is no known algorithm to be applied ahead of time to get the right results.
Again, I am a complete amateur, and I'd love to get constructive corrections or elaborations on this.
Thanks!
-n
The size of a photosite is one of many factors in its sensitivity. Sensitivity dictates the minimum amount of light required to produce information to be recorded.
Dynamic range is the difference between pure black and pure white values that a sensor can read, as well as the number of interim gradations that produce perceptually differential information.
Noise floor is a perceptual thing -- there is some amount of noise in any real-world sensor output. But there is general wisdom about what humans can perceive.
My understanding is that scaling up a sensor's photosite size would not necessarily increase dynamic range, it would only reduce the minimum amount of light required to register a signal, improving its sensitivity. It would also reduce the clip point of the sensor, though, by whatever factor you scaled up the size of the photosites. So if you made the Dragon more sensitive, you would just be clipping skies at a less-bright level, which is not necessarily desirable.
Increasing sensitivity while artificially reducing the output gamma curve in the typically-noisy-region would reduce the perceptual noise floor, since the sensor would be throwing away information that it "knew" it could not reliably capture without unwanted noise. This would also impact the perceived dynamic range, since a smaller amount of information would be recorded at the dark end of the continuum, since it was thrown away because it was probably way too noisy.
It's also my understanding that although these things are, in an academic way, mostly unrelated, in the real world there is a ton of practical overlap and complications, and the real magic is in tweaking these things to get good pragmatic results. This is one of the jobs of the gamma pre-sets that control the debayer process, among many others.
From what I've seen of Dragon footage, it does look like sensitivity is increased, and it also looks like dynamic range is increased. But these things are mostly unrelated. It's hard to tell about the noise floor without more information about what was happening at the time of capture, or by seeing side-by-side footage with another camera that you are familiar with. But you could theoretically increase sensitivity while also making the relative noise floor worse, and vice versa. You could also increase dynamic range, but make sensitivity worse.
MX does not have great sensitivity compared to other digital camera sensors I've used. However, it has great dynamic range compared to other sensors I've used.
I think there was a chart posted that showed Dragon getting 19 stops of theoretical dynamic range, or something. But that doesn't help if, in reality, the bottom 2-3 stops are below the noise floor. It also doesn't help if the darkest stop that the sensor is able to record is as bright as the sun, because then you could never use the camera on planet earth to film stuff. To fix the noise floor, you might have to crush out the bottom 2-3 stops in software, either in firmware on the camera or in the debayer defaults. To fix the minimum sensitivity, you might have to crank the gain from each photosite, which would then make the overall noise floor worse. It's all a bit like a giant 20-sided rubik's cube, except there is no known algorithm to be applied ahead of time to get the right results.
Again, I am a complete amateur, and I'd love to get constructive corrections or elaborations on this.
Thanks!
-n