7D Vertical Banding in Low Light Footage

Started by pest01, February 16, 2014, 02:05:46 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

pest01

Hello everyone i wanted to ask if your having issues with vertical banding in very low light situations on the 7d. It seems to me even when im recording H264 with iso 1600 and a flat picture style it appears very noticeable in the footage.

Walter Schulz

Quote from: pest01 on February 16, 2014, 02:05:46 AMHello everyone i wanted to ask if your having issues with vertical banding in very low light situations on the 7d.

Would be helpful to be linked to a RAW file where this problem occurs.

1%

Yes, 100%.. they are in CR2 shadows too :(

supposedly on 600D/60D its the same sensor but I see 0 banding there.

arrinkiiii

Quote from: pest01 on February 16, 2014, 02:05:46 AM
Hello everyone i wanted to ask if your having issues with vertical banding in very low light situations on the 7d. It seems to me even when im recording H264 with iso 1600 and a flat picture style it appears very noticeable in the footage.


I think raw2dng can solve this issue but mlv2dng not. Maybe in the future the dev's can make an update to the tool.

a1ex

7D has banding in shadows, raw2dng solves it only on highlights (the first one is a black offset per-column, the second one is a non-uniform amplification - each column has a slightly different amplification factor).

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html

(first is 3, second is similar to 5, but per-column rather than per-pixel)

arrinkiiii

Quote from: a1ex on February 16, 2014, 08:52:01 AM
7D has banding in shadows, raw2dng solves it only on highlights (the first one is a black offset per-column, the second one is a non-uniform amplification - each column has a slightly different amplification factor).

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html

(first is 3, second is similar to 5, but per-column rather than per-pixel)

Thank you so much for this link. So, it's possible, at least to getout/attenuate the banding/pattern noise.

"Fixed pattern noise can be removed. By making a template from the average
of 16 identical blackframes and subtracting it from the image most of the fixed pattern noise is removed...
The residual variable component of pattern noise consists in this example largely of horizontal banding noise."

My dumb questions is:

-I take 16 black frame pictures with the cap on but then how i process the MLV file for process the subtracting?

-If i take this 16 photos/blackframes for subtracting and get out the noise pattern, can i use this for other 7D models? Or witch one have there own noise pattern? 

What i can see for a easy solution/workflow is some kind of GUI with/in MLV that you can upload the 16 photos for subtracting, no? 

OSCA LEE

Quote from: arrinkiiii on February 16, 2014, 03:54:30 PM
Thank you so much for this link. So, it's possible, at least to getout/attenuate the banding/pattern noise.

"Fixed pattern noise can be removed. By making a template from the average
of 16 identical blackframes and subtracting it from the image most of the fixed pattern noise is removed...
The residual variable component of pattern noise consists in this example largely of horizontal banding noise."

My dumb questions is:

-I take 16 black frame pictures with the cap on but then how i process the MLV file for process the subtracting?

-If i take this 16 photos/blackframes for subtracting and get out the noise pattern, can i use this for other 7D models? Or witch one have there own noise pattern? 

What i can see for a easy solution/workflow is some kind of GUI with/in MLV that you can upload the 16 photos for subtracting, no?

that's my question as well

Panz

i guess you don't process the mlv file, rather than the dng sequence.

Just layer them in a program(photoshop, resolve, whatever you are using) and choose a blending mode, subtract possibly the best solution.
If that doesn't work try different blending modes till you are satisfied.

Hope it helps

dmilligan

No that's wrong, you do process the mlv file. Dark frame subtraction is not effective after demosaicing, black level correction and nonlinear stretching that occurs when you open a raw file in something like photoshop.

Record a short mlv file with the cap on and exactly the same settings as your clip. You can supply this mlv clip to the mlv2dng as a command line parameter and it will take care of doing the subtraction for you (I can't remember the exact command switch off the top of my head). You can't use different cameras, the banding noise is unique. You can't even use different settings and temperature affects it as well so if your temp changes a lot you probably need to shoot another dark clip.

So in summary: the dark clip must be exactly the same camera, all the same exposure and resolution settings and approximately the same temperature. If any of these change you need to shoot another dark clip.

g3gg0 did an example but I can't find it.


arrinkiiii

Thanks for the info.  So it's possible to make a GUI for upload a MLV file recorded with cap on for make the dark frame for subtraction. That is good news =))

Someone make a simple GUI with MLV_dump with the option for vertical banding for the 5DIII... this options is not good because each  camera need there own dark frame for subtraction (mlv recorded with cap on), right ? Or he use other method? Going try to find the post with this GUI and ask what method he use.

The link that A1ex post they say you need 16 dark frames for making a good average, for video we just need 1 MLV dark frame file?

Record with the same settings, MLV file is good also for that because it got all the data that you need to know for record a MLV file for use as a black frame. 

Thanks again and hope someone make a GUI with this option, for upload a MLV file (black frame) before you hit the button for do the dng's.


EDIT:  This is the link for the app that have the option for fix the vertical banding in the 5DIII  http://www.magiclantern.fm/forum/index.php?topic=10466.msg101344#msg101344



Panz

So it needs to be applied as a pre-processing step to debayering and couldn't be fixed later(is that certain?).

Is the Line Noise Filter in raw therapy able to fix that? It says it fixes pattern noise.
arrinki if you have some footage with this problem can you test rawtherapee solution as well?

I remember having these lines with h264 video but haven't encountered them using mlv_dump lately.
How much do you have to raise the shadows in order for the stripes to be seen?

maybe off-topic, the only problem with this solution is that you need mlv metadata that you don't have after dng extraction(wb,exposure) or is there a way to obtain them from the dngs?

Marsu42

Quote from: 1% on February 16, 2014, 02:48:38 AM
Yes, 100%.. they are in CR2 shadows too :( supposedly on 600D/60D its the same sensor but I see 0 banding there.

No, it's not the exact same sensor - Canon kept tweaking it a bit with every new incarnation starting with 60d, plus of course they recently added the dual pixel af.

The 7d is known for the banding problem, it's not only the oldest version of the 18mp sensor, but the fast & dual readout has a bad effect (the same with 5d3/6d) - it's some analog problem and the severity is reported to vary among specific cameras.

arrinkiiii

Quote from: Panz on February 17, 2014, 10:40:56 PM
arrinki if you have some footage with this problem can you test rawtherapee solution as well?

Thanks, i never used rawtherapee but i will give it a try. I have some raw (not MLV) files that need to be fix because of the vertical banding.

Quote from: Panz on February 17, 2014, 10:40:56 PM
How much do you have to raise the shadows in order for the stripes to be seen?

Don't need to raise or do nothing, it's there. If you use ISO100 or other, but let's say, ISO100 and you shoot some scene that have strong/medium shadows it will appear this vertical banding thing.   

Quote from: Panz on February 17, 2014, 10:40:56 PM
maybe off-topic, the only problem with this solution is that you need mlv metadata that you don't have after dng extraction(wb,exposure) or is there a way to obtain them from the dngs?

What solution? The subtracting black frame (MLV file recorded with the cap of the lent) is process wend you process the MLV 2 DNG's.


It's good to know that exist a solution for this =)) Just hope that soem one can add this to a GUI for be more efficient workflow.

Panz

@arrinki
Well, i haven't see it that pronounced. I usually have to lift the shadows or midtones a little bit for it to appear(at least with h264 and some earlier raw tests).

Quote from: dmilligan on February 17, 2014, 04:21:43 PM
So in summary: the dark clip must be exactly the same camera, all the same exposure and resolution settings and approximately the same temperature. If any of these change you need to shoot another dark clip.
As i understand, in order to apply the black frame correction, you need to shoot with cap on, but with the same settings(lens, f-stop, shutter, white balance,iso). Is that not the case? Or this is pattern noise visible in the same way, in every combination of the above elements?

Also did you have any pattern noise in highlights or midtones or only shadows(that is without using the raw2dng vertical stripe correction fix)?

dmilligan

Quote from: arrinkiiii on February 17, 2014, 09:17:18 PM
The link that A1ex post they say you need 16 dark frames for making a good average, for video we just need 1 MLV dark frame file?
record for at least a second and you've got yourself more than 16 frames ;)

Quote from: Panz on February 18, 2014, 12:33:50 AM
As i understand, in order to apply the black frame correction, you need to shoot with cap on, but with the same settings(lens, f-stop, shutter, white balance,iso). Is that not the case?
That's exactly what I said, changing any exposure setting (mainly Shutter and ISO) is going to change the banding/fpn noise, which will mess up the subtraction and make it much less effective. The actual clip and the dark clip must have exactly the same settings.

Quote from: Panz on February 17, 2014, 10:40:56 PM
So it needs to be applied as a pre-processing step to debayering and couldn't be fixed later(is that certain?).
Yes, ask any astrophotographer. Learn how the math works and you'll see why (read that article a1ex posted thoroughly).

arrinkiiii

Quote from: dmilligan on February 18, 2014, 01:26:12 AM
record for at least a second and you've got yourself more than 16 frames ;)

lol indeed. Just think that the 16 frames will be to process each frame, like a average. But is good to know how we can do it. Thanks =))