5D MK3 Raw sensor ISO noise samples

[a1ex]

Wait a minute, we were talking about video, not photo. They do not behave in the same way.

My samples: 160  200  250

dcraw -h does not do demosaicing, it just groups a RGGB cell in a RGB pixel.

The difference in squig's files comes from different white levels, because I thought raw in movie mode behaves like in photo mode when you change the ISO.

[Audionut]

Wait a minute, we were talking about video, not photo. They do not behave in the same way.

That explains that.

Must remember that camera does video too! 

[IliasG]

Here are 6 samples.  ISO's 160, 200 and 250.  Both with lens cap on, and of a white image as close as I could get to saturation @ ISO 200.

.....

I couldn't really understand the results shown by RawDigger and a1ex's observation that these ISO's are the identical raw data.
So I got my test bunny, and again shot @ ISO 200 as close to saturation as possible.  Then with the same shutter/aperture, I shot again @ ISO's 160 and 250.

I don't really use dcraw so my process might be flawed, but to me, it's clear that ISO 250 has the raw data pushed into saturation.  Based on that observation, ISO 160 simply has all of it's data pushed further south into the noise floor.

So the questions for me, am I simply not using the correct process to recover the detail, or has it indeed been pushed into saturation for the digital pushed ISO's, never to be recovered*?  If I am using the correct process, when in the pipeline is the digital manipulation occurring?  Have the digitally pulled ISO's been pulled into the noise floor of the camera?  Or is it done after the effects of noise etc etc in the circuitry of the camera?
...

*While I can understand that it is (was), the same initial raw data, if it's being pushed into saturation with a resulting loss of detail, it's open to interpretation on whether it's still the same raw data from a usability stand point.

Hello Audionut,

As Alex said we need video-raw files not photo raw. Although your samples will be useful to compare histograms between the two modes.

For a start we need video-raws with burned areas (totally saturated) for every ISO to declare the White clipping point. Then we can continue investigating what happens near saturation.

We will have answers for your questions when we will have proper samples. For the moment just keep in mind that digital push-pull takes place "after the effects of noise etc etc in the circuitry of the camera". By pushing by 1.25 (for ISOs 125-250 ..) it depends on where is the "integer" ISOs real White point, if it's at 0.80 of the max 14bit level then nothing changes ..

[IliasG]

Wait a minute, we were talking about video, not photo. They do not behave in the same way.

My samples: 160  200  250

dcraw -h does not do demosaicing, it just groups a RGGB cell in a RGB pixel.

The difference in squig's files comes from different white levels, because I thought raw in movie mode behaves like in photo mode when you change the ISO.

Hello Alex, thanks for the samples.

As I wrote for Dcraw's -h the interpolation is minimal, just combines the two green channels of a 2X2 square into one. So you loose a little info regarding channel imbalance ..

BTW I think it would be useful for raw2dng and the pattern correction it does if it have available the per channel Black level. Just run 4 loops with N/4 samples instead of 1 with N samples. I think Adobe's converters make use of those four independent values and others will follow .. It can make a difference at high ISOs.

I can change the metadata in squig's samples so this wrong WL was not a problem.

Did you checked Libraw's 4channel code ?.

[a1ex]

Running through the raw samples for vertical banding, collected from the forum, I've noticed the white level is 16382 or 16383. The 16383 one may be because some samples were already corrected for banding, and I've reversed the raw data.

If this is true in LiveView for all ISOs, that's interesting. Will check on the other cameras.
edit: on 5D2 it varies between 15850 and 15882.

I didn't check the 4channel, because the conclusion is clear for me: digital ISO has no effect on LiveView raw data.

I can compute 4 black levels, no big deal. Though it's unlikely to make a difference, can you find some examples?

[squig]

I didn't check the 4channel, because the conclusion is clear for me: digital ISO has no effect on LiveView raw data.

[a1ex]

See the post from Ilias. You need to do the ISO test again btw; the ML version you have used had wrong metadata for white levels.

[IliasG]

Running through the raw samples for vertical banding, collected from the forum, I've noticed the white level is 16382 or 16383. The 16383 one may be because some samples were already corrected for banding, and I've reversed the raw data.

If this is true in LiveView for all ISOs, that's interesting. Will check on the other cameras.
edit: on 5D2 it varies between 15850 and 15882.

I didn't check the 4channel, because the conclusion is clear for me: digital ISO has no effect on LiveView raw data.

I can compute 4 black levels, no big deal. Though it's unlikely to make a difference, can you find some examples?

You are correct, whith the up to date samples there is no difference between ISO 160, 200, 250. May be because of a different LV mode vs squig's (now ancient) samples.

Does White Level goes up to 14bit limit with all ISOs ?. If so 15000 is over conservative, you can set it at 16000 leaving 382 levels clipped just in case a setting makes something different.
Hi ISO NR and long exposures comes in my mind which (in photo mode) results in a different clipping .. instead of a single level it becomes a bell distribution and to prevent magenta casts we have to clip at the lower value of this distribution.
Same as Gulliermo's article .. http://translate.google.com/translate?u=http%3A%2F%2Fwww.guillermoluijk.com%2Ftutorial%2Fsatlevel%2Findex.htm&langpair=es|en&hl=EN&ie=UTF-8

But first we have to check the behavior near saturation (magenta casts).   

I also think that we have somehow corrected data. The sad thinks is that this correction can be erroneous .. sometimes in black frames we see a black band with no noise and I think this is exactly erroneous over-correction.

I could find samples with heavy BL imbalance at high ISOs but only for photo mode which is not useful as we see that LV data are different.

PS. When calculating BL sometimes there are ouliers which disturb the result .. if it's possible to remove them the result would be more accurate.

[a1ex]

Yeah, I'm trying to warn people not to record in photo mode; I've disabled the raw_rec operation in photo mode, and other guys are enabling it back...

Sigh...

For outliers, median rejects them very well, but it's harder to compute than mean. But yeah, a hot pixel in that area may affect the results.

Do you have any samples with black band and no noise?

[IliasG]

..
For outliers, median rejects them very well, but it's harder to compute than mean. But yeah, a hot pixel in that area may affect the results.

Do you have any samples with black band and no noise?

Whats the current state with hot/dead pixels in video raws ?. are they corrected in camera ?.

I have some samples from Canon40D photo-raw black frames with black band  .. I can post when I 'll be back home (weekend) but I don't think they are useful.

[a1ex]

Some cameras set them to 0 (e.g. 60D), others correct them and output clean raw data (5D3). I don't know if there are other situations.

[Danne]

Some cameras set them to 0 (e.g. 60D), others correct them and output clean raw data (5D3). I don't know if there are other situations.

Are you sure that 5d mark 3 "cleans" the rawfile in camera? When in 5x mode I sure get a few dead/hotpixels. I can even see the in playback mode of the rawfile. See posted example. Of course this gets visible in high iso and underexposed shots. Acr is removing them but when I go straight to premiere with the ginger plugin they show up.

[Audionut]

Some updated samples.  Original DNG's

As expected, no difference (slight differences from varying shot to shot read noise) from digital ISO's as compared to their base ISO's.
Massive jump in read noise @ ISO 800.  Read noise then diminishes through ISO 1600 and ISO 3200 where pattern noise starts to dominate.  All noise then becomes amplified through ISO's 6400, 12800 and 25600.

Some resized samples at base ISO's.  +5EV - Temp 3000 - Tint -150

100


200


400


800


1600


3200


6400


12800


25600



A sample @ ISO 800 Photo Mode


The pattern noise in video mode is also destructive as it's not fixed.  It can't be averaged out.
Here is the ISO 800 shot in photo mode with 16 frames averaged to remove the fixed pattern noise.

[eyeland]

So the conclusion is that the digitally pulled/pushed ISO values have no effect on noise in RAW?

[Audionut]

As expected, no difference (slight differences from varying shot to shot read noise) from digital ISO's as compared to their base ISO's.

[heniloux]

Can one conclude that according to AUDIONUTS tests that the BASE STOP of the RAW function would be iso 100? i.e the rating with the least amount of noise?