RAW based exposure feedback

Started by Audionut, May 29, 2014, 05:41:00 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

mothaibaphoto

@Audionut: Please, make this thread "Sticky" !!!

osterizerble

Hi,

Thanks for the hard you guys have put in the development.

I cannot see my raw histogram. When I am in live view it doesn`t seem to show up on screen. It does not show in image review either. When I press info, it only shows me canon histogram. I have 650D and running latest nightly from July, 02, 2015.

Global draw on > all modes
Zebras > Raw RGB
Spotmeter > Raw, AFBOX
Histogram > RGB, Log, ON, Full Histogram, ETTR Hint
Auto ETTR > Always on

How do I activate Raw Histogram to show up on live screen and also in Image review mode.

Thankyou

Walter Schulz

RAW histogram not showing up in liveview is user error. Press Info button several times to bring up ML overlay screen.
RAW histogram not showing in Image review is indeed a bug introduced in Build #367 (24. June 2015):
https://bitbucket.org/hudson/magic-lantern/issues/2339/build-367-2015-june-24-raw-histogram-not

Thanks for testing!

Guillermo Luijk

Would it be possible according to the way Liveview works, to have information in the RAW histogram not only about clipping taking place, but about how many stops of highlight info are currently getting clipped? i.e. an estimation of the histogram of clipped RAW data.

E.g. the orange bars represent info that would get lost if we click:



With Liveview capture working 2 stops below user settings (either with lower capture ISO or exposure), ML could plot a RAW histogram including those 2 (or any other number) extra stops of clipped RAW data with present user settings. This could visually and intuitively help in achieving accurate ETTR in M mode.

I actually prefer zebras over the histogram, but this would provide an enhanced histogram anyway.

Regards

a1ex

I've been thinking about this lately as well. Currently, ETTR does some guesswork on this based on past histogram data (past frames), but there's no reason why this estimation couldn't be plotted in the regular histogram.

Walter Schulz


AOK

Hi everyone.

The best article/post on the subject so far. Thank you very much for taking the time posting it. :-)

Question: I understand that the analyzed data e.g. DR, histogram, ETTR goal, etc. are displayed based on RAW. However the image itself when shot - is its jpg preview as in the regular firmware or it's the RAW file?

I am asking because as bad as the camera display could be, still one could better judge the results if it's (trying to represent at its best) the raw actual image and NOT the discrete contrast and compressed jpg version. The feedback would be great.

So is ML displaying the RAW image or (its closest representation) or it's also dealing with bad jpg version of the image?

Thank you in advance for your answer and excuse my limited knowledge on the subject.


dmilligan

The display itself is low resolution, 8 bit/channel, YUV 422. There's no doubt it would be impossible to tell the difference. All the more reason you should never judge anything about exposure based on the way the image looks on the screen. Always rely on exposure aids like histogram and zebra.

AOK

True. :-) But why would be there such huge fallback in all firmwares? I mean since ML does it, then it's possible to view histogram and zebra highlights based on RAW and not on JPG.

Why with firmwares it's always JPG analysis and representation? Battery drain, image processor (most probably not)? Why do they keep using the embedded JPG?

Roberto Mena

I need clarification on how to use the ETTR tool with the RAW Histogram on ML when shooting video. When the ETTR tool reads EV 0.0 that's a perfect exposure being read by the ETTR tool, correct? But if I really want to do ETTR I should then open the camera iris for it to read a negative value correct, like EV -1.0 that way in post I can lower the exposure, etc. to get a better image with more data, correct?

dmilligan

If you expose too far to the right then you will clip parts of the image you probably don't want to and actually loose information. Exposure is like "The Price is Right", you must get as close as you can without going over.

Audionut

Use zebras.  Then you can keep pushing the exposure until wanted detail is ETTR.

The histogram will only tell you that some part of the image is being over-exposed, not specifically which part of the image.

garry23

@Roberta Mena

From my experience, as a stills photographer and not a videographer, I believe some get 'confused' by the additional functionality provided by the mid and shadow SNR.

The SNRs will/may push your image over into the highlights being overexposed, because these setting are saying to ETTR, 'please set a highlight ETTR, BUT, if my shadow (or mid) SNR criterion is not met, then 'sacrifice' the highlights.

You cant' have your cake and eat it  ;)

For this reason, personally, I set the SNRs to zero, and 'just' use the shutter speed, exposure target and highlight ignore, plus RAW Zebras. ETTR+RAW-Zebras+RAW-Histo-Ev-Hint is a very powerful combination, giving me the control, rather than the camera (if you rely on the SNRs).

Cheers

Garry

Roberto Mena

Thanks for all the responses but I pretty much knew most of the things I was told on here. FYI, I have been using the ETTR tool for shooting video and I do usually over expose by -1.0 EV and so far the images I've gotten are good. I might starting knocking it down to -0.5 EV to be more of the safe side. My confusion comes from the values of 0.0 EV on the ETTR tool. In other words, when it does read 0.0 EV that means it is already overexposed to the right and now its a matter of how many more stops to push it? Or does 0.0 EV read as an already properly exposed image and now I need to start opening the iris more to do ETTR?

garry23

0Ev means no highlight clipping. Full stop.

Thus any speculars will control/dominate your exposure.

a1ex

Quote from: garry23 on January 13, 2017, 12:16:26 AM
0Ev means no highlight clipping. Full stop.

Wrong answer. This is only valid when highlight ignore is set to 0 and SNR limits are disabled.

The exact meaning of 0.0 is given by the options you have set in ETTR submenu. It can range from very underexposed to severely clipped, depending on your settings and your subject.

http://wiki.magiclantern.fm/ettr
https://www.magiclantern.fm/forum/index.php?topic=5693.0
https://www.magiclantern.fm/forum/index.php?topic=17994
https://www.magiclantern.fm/forum/index.php?topic=18024
https://www.magiclantern.fm/forum/index.php?topic=5200.msg114522#msg114522

Roberto Mena

a1ex @ "The exact meaning of 0.0 is given by the options you have set in ETTR submenu. It can range from very underexposed to severely clipped, depending on your settings and your subject." Cool but does that mean that the default settings in ML's ETTR the value of 0.0 EV is already ETTR with a little wiggle room for safety? This according to the last URL you posted which has four photos of a white flower, first two using Canon's meter and the last two using ML's ETTR. As mentioned, I think this might be true because I have been shooting my scenes at -1.0 EV and the images did not look overexposed. I also imagine since my 7D's dynamic range is only 11.7 stops it would make sense that I could only over expose one stop anyways unlike the BMPPC that has 13 stops of DR and probably could get away with even overexposing 2 stops to do ETTR... not that I would do that, just saying if that happened by accident the image would still be recoverable in post with curves. Right?

Audionut

Quote from: Roberto Mena on January 12, 2017, 07:42:39 PM
I do usually over expose by -1.0 EV and so far the images I've gotten are good. I might starting knocking it down to -0.5 EV to be more of the safe side.

Are you talking about the "exposure target"?

QuoteWhere to place the highlights with respect to overexposure.

By setting this value to -2.0 EV, you are telling ETTR to place the highlights 2.0 EV (two stops) away from overexposure.  By setting this value to -0.5 EV, you are telling ETTR to place the highlights 0.5 EV (half a stop) away from overexposure.

If you want to maximize the signal to noise ratio of your images, you want to place the highlights as close as possible to 0EV (saturation / clipping / overexposure)

Quote from: a1ex on January 13, 2017, 12:25:13 AM
The exact meaning of 0.0

In the histogram?


garry23

A1ex

Sorry my 0ev answer went with my earlier answer saying I set SNRs to zero.

Garry

a1ex

Got it.

Even so, with a nonzero highlight ignore setting, you will get some clipped highlights (the percentage you have selected, as area or number of pixels). So it's still a bit different than "No highlight clipping. Full stop.".

In other words, with a reading of 0.0 and SNR limits off, if Highlight ignore is set to 1%, that means 99% of the pixels will be below the target level (Exposure target), and the remaining 1% will be above. If there are no specular highlights, there won't be anything clipped. If you do have specular highlights and their area is smaller than 1% of the total image, they will be clipped, regardless of how bright they are.

With highlight ignore set to 0, and SNR limits still off, any strong specular highlight (or even some hot pixels, if unlucky) will dominate your exposure, even if most of your image ends up black.

This is where the SNR limits come in: if the shadows and/or midtones become too dark, enabling those limits will sacrifice the highlights to protect the shadows.

garry23

Understood  ;)

I will not be so lazy with my attempts at helping next time  ;D

Cheers

Garry

Audionut


garry23

@Audionut

Thanks for refreshing things.

One thing I've noticed is that when using dual-ISO with RAW zebras, I don't see the diagonal stripes that you show in the OP.

I'm using an EOSM with lua-fix build from July.

a1ex

If you upload a CR2, I'm able to re-create the overlays in QEMU (as they were displayed in the camera using any given build).

garry23

@A1ex

Here is a test image and .xmp.

Taken with latest (sic) experimental Lua fix.

https://www.dropbox.com/sh/1s82dhko4k1jqvr/AADH9Gb2qvoVi-5XL33PmqT3a?dl=0

Cheers

Garry