EVA1: Footage seems overly bright when ETTR, even when not blown?

Just went back to my GH5 to look closer at how I've been shooting on the EV index and what it generally seems to be about... +2 stops over... on ETTR. Blowing out only some useless highlights and keeping everything else as bright as possible.

When I expose to "0" on the EV it looks stupid under-exposed to my eye on the screen... though that could just be what I've been used to... and more centered on the histo/WFM.

I feel like I am blowing my mind and re-learning, unlearning, everything I've been doing. Since I've been happy with my footage before, and now I'm seeing the footage respond totally different on EVA1, I've realized it's a bit more complex.

Frankly, I agree with the sentiment of "expose correctly" but... Mind blown. I don't know what "correctly" usually is now... or why correct is correct... other than what looks right to my eye.

If i had to guess / summarize:

Everything as close to middle gray as possible, with only outliers towards the shadows/highlights. On GHx favor ETTR so everything is brighter; on EVA1, favor pushing things further down because it responds better in the shadows with more detail and less noise?

But then, on the GH5/S with vlog, it's also true that in the curve there's less data up high... so.. maybe I should stop utilizing ETTR. Didn't Paul Leeming propogate always ETTR with his LUTs? That was for 10-bit vlogL.

Not trying to over simplify, but frankly, I'm not going to be running around in India as a OMB with a light meter shooting live content as it unfolds.
 
Easy way to expose correctly on the EVA. Use the center spot meter in VLOG and will show a under over indicator on the screen in stops. I usually keep it 0 to +1 stop range for my subject. If your LCD is dark, use the power LCD function to brighten it.

Also shoot some stuff purposely under, normal and over exposed. Hookup your camera to a TV (hopefully calibrated) and review the material using the built in LUT to get a feeling for how the EVA responds.
 
filmguy123 said:
since most data on the logarithmic curve is allocated to the lows and center
Click to expand...
That's just not true. On a logarithmic curve, the idea is to allocate exactly the same amount of bits to each and every stop (insomuch as possible). In practical reality the lowest/darkest stops don't get the equal allocation, but once you get past those first few dark stops, every stop after that is allocated approximately the same number shades.

In a linear sensor, recording linear data, the situation is far more exaggerated, but still the opposite of what you're describing. In a linear sensor, each brighter f-stop is allocated more shades than every stop that comes before it combined. So (simplified explanation) the first stop gets 1 bit, the second stop uses 2 bits, the third stop uses 3 bits, fourth stop uses four bits, fifth stop uses five bits, etc. That means:
stop 1 = 1 shade
stop 2 = 2 shades
stop 3 = 4 shades
stop 4 = 8 shades
stop 5 = 16 shades
stop 6 = 32 shades
stop 7 = 64 shades
stop 8 = 128 shades
stop 9 = 256 shades
stop 10 = 512 shades
stop 11 = 1024 shades
stop 12 = 2048 shades
stop 13 = 4096 shades
stop 14 = 8192 shades.

That's the insanity of linear encoding. That last f-stop takes up literally more than half the entire data rate of the entire signal. And the first five stops are encoded with paltry allocations. And you would need 14 bits of data to encode that signal.

That's one reason why ETTR was developed -- because it pushes the image up into the area where the most bits are allocated. I think, in general, it's a tad silly way to shoot, as it forces extreme color correction to be applied to every single shot, and you will have wildly fluctuating tones from shot to shot. But, it did take advantage of the way linear encoding works. And, raising the image up higher in the tonal range results in a higher signal-to-noise ratio, since the darker stops have fewer actual "signal" bits, so an equivalent number of "noise" elements disproportionately will affect the darkest stops.

But, again, linear is a silly way to encode sensor data, as it's profoundly wasteful. There is no way any rational person could discern 8,192 shades in one f-stop of light! It's egregious overkill.

With logarithmic encoding, especially as implemented in V-Log, the first four stops are encoded in 64 shades total. After that, pretty much every stop (so the 10 brightest stops) all get about 64 shades each. That's not strictly mathematically exact, but it's close enough to get the concept across.

That's the whole point of log -- to give you encoding in a reasonable data rate that preserves plenty of shading across the sensor's entire range. With log encoding, V-Log preserves 64 shades per stop, while only taking 10 bits. If you wanted to preserve a 14-stop sensor's data using linear encoding, it would require at least 14 bits, and possibly more depending on how many bits you allocate to the darkest shade; if you started the darkest shade with 3 bits (to encode 8 shades) then by the time you hit stop 14 you'd need a full 16 bits of data to encode the entire data path.

Now, keep in mind that 64 shades per stop is a LOT. Consider that in an 8-bit codec, you'd be looking at a maximum of maybe 8 shades per stop.

So log-encoded 10-bits will give you sufficient practical data in the brighter stops, and more practical data in the lower stops than 14-bit linear would, even though linear would increase the required data rate by 40%.

None of this is really what you were asking for, but hopefully this addresses how sensors encode data and where the data is allocated on the exposure range, and why "ettr" is not really all that important when recording with a log-encoded gamma. ETTR is not inherently "bad", it's just -- not all that necessary, and it does introduce its own set of complications. I think for most purposes most shooters would be better served to follow Soarprod's advice above.
 
Last edited:
Great stuff Barry (and many others here). Someone reminded me of this article (with pix of histograms!) by Art Adams from a few years ago. Some things have changed since then (mainly, RED's and BMD's produce much better images these days), but still some good background and visuals explaining what's going on and why, well, take it away Art:


Why “Exposing to the Right” is often Completely Wrong

Still photographers love histograms. Cinematographers hate them.
Here’s why our world is so different from theirs in this regard…

http://www.dvinfo.net/article/acquisition/why-exposing-to-the-right-is-often-completely-wrong.html
 
What is the vlog IRE cap suppose to be on the WFM on EVA1 when viewing as vlog? 80 IRE, 85, or 100, or 105...?

Does this differ when outputting to an external monitor with no LUT and viewing it's waveform?

What about with a Rec709 LUT applied?

I am having trouble trying to determine what to trust on WFM and Lut application.

On the smallHD when I view as vlog I can expose much higher... it appear to 85 IRE. When I put a rec709 lut, the image changes and things not over-exposed when viewing without the LUT now appear over exposed, as if already clipped.

Why, and how does these interact?

Sorry for the questions, it's still a bit confusing to understand how these are interacting especially when viewing through a monitor in addition and attempting to reconcile the different IRE clip points.
 
Not sure about other questions but I think the EVA is around 92ire.

filmguy123 said:
What is the vlog IRE cap suppose to be on the WFM on EVA1 when viewing as vlog? 80 IRE, 85, or 100, or 105...?

Does this differ when outputting to an external monitor with no LUT and viewing it's waveform?

What about with a Rec709 LUT applied?

I am having trouble trying to determine what to trust on WFM and Lut application.

On the smallHD when I view as vlog I can expose much higher... it appear to 85 IRE. When I put a rec709 lut, the image changes and things not over-exposed when viewing without the LUT now appear over exposed, as if already clipped.

Why, and how does these interact?

Sorry for the questions, it's still a bit confusing to understand how these are interacting especially when viewing through a monitor in addition and attempting to reconcile the different IRE clip points.
Click to expand...
 
filmguy123 said:
What is the vlog IRE cap suppose to be on the WFM on EVA1 when viewing as vlog? 80 IRE, 85, or 100, or 105...?
Click to expand...

V-Log, per spec, should be shot at around 60 ire (I think technically, it's 61 or 62, but how can you really tell on that tiny little waveform?). Now, does that mean that you get hard clipping as soon as you push past that? Not necessarily, but you start losing detail, chroma starts over saturating, and a few other things start getting out of whack (as well as reduction in DR!) as you increase your exposure and really start to see things actually blow out. I've comfortably gone as far as around 70 ire on highlights that weren't particularly important, but YMMV. The only argument to make for shooting above 60 is for noise reduction (as in ETTR), but again, there are compromises to make.

filmguy123 said:
Does this differ when outputting to an external monitor with no LUT and viewing it's waveform? What about with a Rec709 LUT applied?
Click to expand...

If you are recording to V-Log, it doesn't matter if you are sending V-Log or V709 to the LCD... the waveform monitor is still going to reflect the levels that your V-Log footage is being recorded at. This is why it is important to expose using your waveform/zebras, and not your eyeballs, especially when using the V709 M-LUT in the LCD. This is also true of external monitors/recorders where you are sending V-Log signal. Even if you throw on an M-LUT to view it in rec709, your waveform/zebras/vector is still going to show you the V-Log levels... which is ideally what you want anyway! On the other hand, if you were to send V709 (instead of V-Log) from your camera to the monitor/recorder, then the scopes will be showing the levels from the M-LUT, because now it is baked in.


filmguy123 said:
On the smallHD when I view as vlog I can expose much higher... it appear to 85 IRE. When I put a rec709 lut, the image changes and things not over-exposed when viewing without the LUT now appear over exposed, as if already clipped. Why, and how does these interact?
Click to expand...

V-Log, when viewed on a rec709 monitor (without a rec709 LUT) appears very low contrast and saturation, obviously. Just because you don't see the clipping, doesn't mean that it won't be there in post after you transform it. There is a ton of headroom in the video signal to allow for the transform later (and if you notice, you can't even really push it all the way up to 100). I think that your confusion might be based on a habit of exposing with your eyes instead of your scopes/zebras? If that is the case, unlearn, unlearn, unlearn! :)
 
Last edited:
Thanks SeaniK!

Whoa, 60? Where did that come from? I thought that would be a ideally midrange, not clipping point. What is 80ish then? Pure white, with extreme highlight reserved for 60-80, skin tones around 45-50, and most bright spots ending around 60?

One thing I am confused on is that your post seems to contradict my smallHD experience, but I am probably just a bit confused. It would make sense to me if the waveforms and zebras remained consistent and things only looked different to my eyeballs. BUT, on the smallHD, when I am on vlog view with no M-LUT (guessing here but I assume M-LUT means "monitor LUT"?) the image indeeds appears flat as expected, no zebras show, waveform sits further down. As soon as I tap the "look" on on smallHD (no recorder here just monitor on SmallHD Focus) the image changes to punchy along with the zebras, which now appear on bright spots, and the whole waveform shifts upward to become brighter.

I would have thought my scopes/zebras should remain consistent, and things only look different to my eyeball?
 
Obviously then the Small HD WFM is monitoring the transformed image, post monitor lut (M-LUT). Sorry about that, I thought that monitor gave you a choice for what the scopes reflect (default being the actual video being input to it). No wonder that was confusing!
 
I need to look into the options on this thing. I've got one too many pieces of gear at once and so it's quite a cluster. Atomos, SmallHD, GH5S, EVA1 and my head is spinning, but I'm getting there!

I would have assumed it only monitored the signal and didn't have the scopes reflect the LUT. Why would I ever want the scopes to reflect not the actual recorded data?

Also I'm still a little lost as to how the curve expansion happens (or doesn't happen?) later with vlog... when a LUT is applied in post, do all the signals and values stretch back out from 0-100? And why is vlog capping @ 60 when I could've swore I heard 80 before?
 
filmguy123 said:
Just went back to my GH5 to look closer at how I've been shooting on the EV index and what it generally seems to be about... +2 stops over... on ETTR.
Click to expand...
Assuming the EV index is not confused by things like back lighting I think exposing 2 stops over is asking for problems.
 
You must log in or register to reply here.
Back
Top