Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

noise..... what's wrong

OH, didn't realize you mean 2k@120.

I was like, why would 2K be more compressed than 4K (Assuming same frame rate) :)
 
:)

Although, even at 2k@24fps, 2k is MORE compressed than 4k. I couldn't even begin to know the inside and outs of how the codec data-rate changes between resolutions and framerates. All I do know is that I've seen more 'compression-based' artifacts on 2k footage than on 4k footage center cropped to 2k.
 
Hey Vush. I looked at your outdoor example. I feel your pain. I got my RED in November and its been grey low contrast days every time I want to go outside and shoot. Feed the sensor light and your noise problem will diminish significantly. On low contrast days try to push the histogram as far to the right as you can without clipping. You can pull the highlights down in post. I don't see any reason to shoot 500 ISO on a day like that. If you can, try to keep it between 250 and 320. That is where most people are rating the camera anyway.

As for comparing to the HVX noise, in my experience, the RED is worlds apart in that department. The noise the HVX would produce in gradation between light and dark seemed exaggerated at times. I don't see that problem at all with the RED, so even if you have to crush the blacks you shouldn't run into the problem of fighting off yet more noise in the mid-tones. I think the HVX is a great camera by the way. I can't say enough good about, but, yeah your experience with RED should be significantly better.

Good luck, keep us up to speed on your results.
 
The footage doesn't look too bad to me. I wouldn't be worried, it seems normal.

If anyone starts getting vertical lines (of noise), or dithered patches of pixels that don't look right, you might have a loose cable to your sensor (like i did). Noise characteristics got worse over time, but should get camera back after the weekend.

vushvush said:
OH, didn't realize you mean 2k@120.

I was like, why would 2K be more compressed than 4K (Assuming same frame rate) :)
Click to expand...

I wouldn't expect a 2k image to compress as small (av. data per pixel etc) as a 4k image.

Neither would I necessarily expect 2k to compress AS NICELY as a similar 4k shot, since larger spatial images tend to compress better, possibly giving the codec more room to work its magic.

That said, I don't know how adaptive redcode is, but it has to compress a lot of data, and in realtime.

Redcode converts image data into wavelets before compressing them.

A "Wavelet" is basically two numbers, a "difference" coefficient and a "signal" coefficient, calculated from two or more pixels/signal samples .

In theory an image can be broken down and represented with a single "signal" coefficient (just a number) and lots of "difference" coefficients (the number of these depends on the original resolution), though in practice images don't tend to be broken down this far.

During decompression, these "difference" coefficients are fed back into any remaining (after compression) "signal" coefficient(s) to recursively reproduce higher and higher resolution arrays of signal coefficients (images), until all the difference coefficients have been used (unless your doing a half resolution decode etc for previewing, in which case you don't reproduce the complete image (signal), or use all the "d" coefficients), and the image is decoded in full.

The "difference" coefficients produced as the CFA is split into wavelet coefficients (at smaller and smaller resolutions) tend to have smaller values (unless the image is mostly noise) as spatial resolution is increased (since they are literally just numbers representing the differences between adjacent pixels), and subsequently can generally be compressed better with whatever compression system is used (ie quantization & RLE, arithmetic coding, huffman coding etc etc).

NB, wavelets are WHAT is being compressed, they are not the compression system itself, which can be one of many systems as mentioned above.

The reason the R, G and B samples (taken from the sensor/CFA) are converted into wavelet coefficients in the first place, is to analyse the original signals (image) and identify "less important" data that can be thrown away (and enable visually lossless compression with low data rates, whilst minimizing artifacts)

Wavelets have the advantage of being fairly fast/efficient to compute, especially compared to traditional codecs such as JPEG/MJPEG/DV, which have to convert blocks of image data from the "time/space domain" into data representing the "frequency domain" (don't ask) using pretty complex/slow math, in order to work out what can be quantized, where, and how much.

Because these calculations aren't performed on the whole image (not least because of computation times), the image has to be broken into smaller blocks, and this tends to lead to visible artifacts after compression has done its thing. Wavelets on the other hand work in the "time/space" domain (don't ask again) directly, which means less computing.

Wavelets also have the advantage that the information thrown away tends to be distributed throughout the image, and so macro block artifacts aren't present, as well as being fairly efficient to implement (in terms of simplicity, memory usage and speed)

There are lots of different wavelet analysis methods, I won't pretent to know exactly how they all work. I've only read/implemented whats known as a HAAR function (very simple, but useful in understanding wavelets and some of their different uses).

If you know more than I do (not too hard), please feel free to correct me
 
vushvush said:
outdoor shot:

http://www.makevisual.com/download/A001_C009_120163_001.R3D


As for the color, it shouldn't matter since the color balance and REAL exposure (with the ASA) is done in post. I usually keep things between 160-500.
Click to expand...

I see what you are saying here but color balance will affect your image greatly by view. You can get rid of noise through color management, so the reverse is true, if you have improper white balance you can get a lot of noise... Now you can change that later on but it's no use freaking out if you know the noise will disappear. It's best just to set the proper white balance if you can...

Leaving asa on 320, the sensors native position, will see you in the right place for exposure. Light and expose for that as a starting point and make compromises from there when you have to...I look at asa as gain. It never leaves 320 on my camera.

I saw two movies in last month, James bond and Australia and both exhibited noise that would eek me out if i was in image control. Now thats either a new phenomena (because i have never noticed it before) or i am sensitive to the reduser noise and have therefore started to notice what was always there...


I think you can maybe focus on it to much sometimes. There's a noise level that is obviously acceptable to audiences. Be careful not to see the forest through the trees and get lost in the small detail...
 
I try to keep things simple this way:
you get the best results at 4k, 320 asa (or less) and 5000degrees kelvin. That is what the environment needs to give the camera, so if you don't have enough light you need extra lights or a faster lens, if the color temp. is not okay you need a filter for the lens. This may give you some extra work before shooting, but saves you a lot of work afterwards
 
ronx said:
The footage doesn't look too bad to me. I wouldn't be worried, it seems normal.

If anyone starts getting vertical lines (of noise), or dithered patches of pixels that don't look right, you might have a loose cable to your sensor (like i did). Noise characteristics got worse over time, but should get camera back after the weekend.



I wouldn't expect a 2k image to compress as small (av. data per pixel etc) as a 4k image.

Neither would I necessarily expect 2k to compress AS NICELY as a similar 4k shot, since larger spatial images tend to compress better, possibly giving the codec more room to work its magic.

That said, I don't know how adaptive redcode is, but it has to compress a lot of data, and in realtime.

Redcode converts image data into wavelets before compressing them.

A "Wavelet" is basically two numbers, a "difference" coefficient and a "signal" coefficient, calculated from two or more pixels/signal samples .

In theory an image can be broken down and represented with a single "signal" coefficient (just a number) and lots of "difference" coefficients (the number of these depends on the original resolution), though in practice images don't tend to be broken down this far.

During decompression, these "difference" coefficients are fed back into any remaining (after compression) "signal" coefficient(s) to recursively reproduce higher and higher resolution arrays of signal coefficients (images), until all the difference coefficients have been used (unless your doing a half resolution decode etc for previewing, in which case you don't reproduce the complete image (signal), or use all the "d" coefficients), and the image is decoded in full.

The "difference" coefficients produced as the CFA is split into wavelet coefficients (at smaller and smaller resolutions) tend to have smaller values (unless the image is mostly noise) as spatial resolution is increased (since they are literally just numbers representing the differences between adjacent pixels), and subsequently can generally be compressed better with whatever compression system is used (ie quantization & RLE, arithmetic coding, huffman coding etc etc).

NB, wavelets are WHAT is being compressed, they are not the compression system itself, which can be one of many systems as mentioned above.

The reason the R, G and B samples (taken from the sensor/CFA) are converted into wavelet coefficients in the first place, is to analyse the original signals (image) and identify "less important" data that can be thrown away (and enable visually lossless compression with low data rates, whilst minimizing artifacts)

Wavelets have the advantage of being fairly fast/efficient to compute, especially compared to traditional codecs such as JPEG/MJPEG/DV, which have to convert blocks of image data from the "time/space domain" into data representing the "frequency domain" (don't ask) using pretty complex/slow math, in order to work out what can be quantized, where, and how much.

Because these calculations aren't performed on the whole image (not least because of computation times), the image has to be broken into smaller blocks, and this tends to lead to visible artifacts after compression has done its thing. Wavelets on the other hand work in the "time/space" domain (don't ask again) directly, which means less computing.

Wavelets also have the advantage that the information thrown away tends to be distributed throughout the image, and so macro block artifacts aren't present, as well as being fairly efficient to implement (in terms of simplicity, memory usage and speed)

There are lots of different wavelet analysis methods, I won't pretent to know exactly how they all work. I've only read/implemented whats known as a HAAR function (very simple, but useful in understanding wavelets and some of their different uses).

If you know more than I do (not too hard), please feel free to correct me
Click to expand...


I press the little red button....:usd:
 
So other than light exposure, there is nothing else I can be doing to lead to the noise?

And more specifically, what I think I'm hearing is that the noise in the outdoor shot is typical?
 
Noise is also a brain ghost. The more you look for it, the more you'll find it. :)

To get minimum noise:
- Rate camera 320 iso or lower.
- Maximise exposesure without clipping white
- Use dalight, 5000 kelvin
- Crush black in post.

..but noise it not same as evil. Look at any feature film, and you'll find tons of noise.

Cheers
 
Can one of you post a daylight outdoor shot similar to the one I shot so I can see what you would consider good?

(.r3d would be ideal)
 
Vush,
Your outdoor image is exposed fine. Any brighter and those silver cars will blow out. It didn't seem overtly noisy to me. Remember 2k is much noisier than 4k. Try taking some 4k footage at 50mm and then the same image at the same stop/shutter speed at 25mm 2k. Scale them both to 1080 and look at the difference.

The sink shot is underexposed with low contrast. That'll be noisy every time.

If you convert the outdoor footage with redspace for both settings there shouldn't be any noise. Also, if you are just looking at 1/4 res medium playback in redcine, you'll be seeing more noise than in the full debayered transcode.
 
Did a quick test with the outdoor shoot.
There isn't much noise in there, when developed properly.
2k is an issue, though. But a noise-issue.
 
RC
From my head (somewhere else now)
K 5000 (or 5600)
Tint/hue : 0
ISO/ASA 250 (or 320 with -0.4 exposure), just lowering exposure so the cars don't blow,
some WB in matrix. I may have hit 160...

Some quick curve/brightness/contrast stuff

If I was to grade, I'd have done cam RGB/RedLog
For this test, I just stuck to redspace.

No de-noise/sharpening.

Your video was very blue. I could recreate that in the matrix, if that was what you wanted.

Just pull down the red and green sliders in the matrix, don't gain the blue (you can get any hue by changing any pair of two sliders)

sorta (from my memory)

Tell me if you're happy.

Cheers!
 
JanneJansson said:
Noise is also a brain ghost. The more you look for it, the more you'll find it. :)

To get minimum noise:
- Rate camera 320 iso or lower.
- Maximise exposesure without clipping white
- Use dalight, 5000 kelvin
- Crush black in post.

..but noise it not same as evil. Look at any feature film, and you'll find tons of noise.

Cheers
Click to expand...

I agree with everything here excpet the use of 5000 kelvin. Does it matter where the kelvin is set or is it just a perception thing? Kelvin is metadata so I'm not sure that setting effects the amount of light hitting the sensor, or am I missing something?
 
I think what Janne is referring to is ideal conditions. RED works best when daylight is present.
 
vushvush said:
I think what Janne is referring to is ideal conditions. RED works best when daylight is present.
Click to expand...

Ouch... If you're implying that tngsten is a real problem, I have to say that I don't agree. If you imply that daylight is better, well then... sure... :)
 
You must log in or register to reply here.
Back
Top