Bars facelift

[Audionut]

- half-height "dark tail'
- midpoint indicator
- 1/2 stop resolution (11 stops shown)
- compressed quantitative "clip" indicators

IMO, you've made it to complicated stevefal.  With the DR indicator showing 11 stops with 11 bars, it's easy to see the what each bar represents.  Also, I think 1 bar = 1EV seems the most appropriate, anything else seems counter-intuitive.

If the highlights are 3.5EV below saturation, with this implementation I would have to count 7 empty bars to be sure how far I am underexposed.
Using half (full) bars I can see very quickly that the top 3 stops are empty and the forth stop is half full.  There is less information to process.  Also, If I don't care about being very accurate, I know (very quickly) I can raise exposure 3 stops and be there or there abouts.
I have suggested to a!ex to use shading in the shadow areas also (rather then 1/3 bars).

[stevefal]

Of course the same design could be drawn in 1 stop resolution - 11 lights. I personally like the idea of 1/2 stop increments when my Av/TV hardware is set to 1/2 stop per click. One click = one light.

I think you are saying you'd prefer 11 boxes, and each box fills in two steps - horizontally - right? It's essentially what I've drawn with every other dividing line removed.

I wouldn't exactly say that's less complicated than what I've drawn, because it introduces two different step types - half steps and whole steps, versus "steps" (or clicks, for me). But it would map better to the way exposure is discussed numerically. I personally do things visually/graphically, so I don't think in terms of "3.5EV below saturation." That's what I figure the numeric ETTR indicator is for.

I'm interested in other feedback on this too. Thanks audionut.

[Audionut]

What if the user is in 1/3rd stops?
With 1 bar = 1EV, you (should) know based on your exposure level increments how many clicks are needed.  2 clicks for half stops, 3 clicks for third stops.

With twice the number of boxes representing the same information, I don't see how it is anything but more complicated. 

Boxes filled in 2 steps vertically would be a better option imo.

I wouldn't exactly say that's less complicated than what I've drawn, because it introduces two different step types - half steps and whole steps, versus "steps" (or clicks, for me).

Because you have the option to ignore the half steps.  You can simply count the empty boxes.  Where the exposure is 3 stops under, there are only 3 boxes to count, not 6.   With 1 bar = 1EV, there is always half the number of boxes to count vs 1 bar = 0.5EV.  Half filled boxes provide the extra resolution only where needed.

[stevefal]

I could personally go along with any variety of ways regarding number of boxes and how to draw them. But the idea of filling them alternately vertically and horizontally makes no graphical sense. If the horizontal axis connotes light in stops/substops, then subdivisions of light should absolutely be oriented horizontally. There's no way that a user could intuitively infer that the vertical axis *also* represents stops.

Think about histobar like this: A traditional exposure histogram represents distribution of light with maximum horizontal resolution (value) and vertical resolution (quantity). As such it provides a detailed picture of the distribution. But it has drawbacks: 1) distributions often have skinny tails and peaks that, although important, are had to see; 2) the relationship between light distribution and camera stops is not always apparent; and 3) scene content that the photographer considers outside the dynamic range of the camera are nonetheless squeezed into the visualization of what's "in" the picture.

Histobar essentially quantizes both dimensions of the histogram so that 1) skinny tails has as much visual weight as fat bellies, 2) level distributions are binned in stops, 1/2 stops etc. and 3) out of range values are accounted for outside the normal values (in the latest concept).

The extreme vertical quantization is like saying to the user, "There are enough pixels at this level that you are going to try to expose them properly." With a1ex's request for demoting the lower 5th percentile, it says, "There are less worthy pixels here (due to likely noise), so try to expose them, but sacrifice these first"

So in the the latest concept, the vertical axis represents the worthiness of the pixels. There are two worthiness values - high and low.

I agree that horizontal steps should be designed so the user can intuitively and quickly dial in exposure. Whereas the latest concept has 22 steps, I can try to draw an 11x2 and/or 11x3 approach too. But I'd like to hear from more people first.

[Audionut]

a1ex currently has them 1/3rd filled for shadow areas that are under some 5th percentile thingy.  No user could ever possible see that as intuitive.  My first response when seeing that, "why are the bottom stops only 1/3rd full".
Hence why I believe a lighter shading of the clipped shadow indicator would be preferred.  In this way, I have a direct visual clue that those (lighter shaded) areas must have some representation of the shadow areas.  I can use different shades of blue to determine that.  I'm smart enough for that!


The vertical distribution in a histobar has no meaning other then the one we apply.

I imply to use the vertical axis, as the graphical representation will be easier to see at a glance then a horizontal axis distribution.

I would go further to say that the vertical distribution in the bar (bucket) could be directly related to the EV difference from the next stop.  ie:  Where the exposure is 0.7EV from the next stop, the bar could be filled 30%.  However, half stop resolution is probably sufficient.

My dumb brain finds it much easier to match the distribution of the vertical axis with exposure, rather then some weighting of priority based on some percentile.
Bar is full, I am at that exposure stop.  Bar is half full (regardless of axis), exposure is half of that stop.

One should always assume that the user will not rtfm.  That's not to say that my ideas are best.

[stevefal]

The vertical distribution in a histobar has no meaning other then the one we apply.

Hah, are you trolling me? I don't like to be pedantic (ok I do), but that's wrong. In UI design, everything has the meaning that the user applies. When the meaning that the user applies is the meaning we implemented, then it is intuitive. That isn't to say that any of these are are necessarily intuitive, but we should be able to agree what the word means. I'm game to discuss the science behind this, but we probably shouldn't here.

I'll try another concept that minimizes horizontal quantization, and uses a texture, rather than short lights, for the 'noisy tail'. This will address both of your issues in another way.

[tonybeccar]

I agree with Doyl4, it'll be very functional to have indicators for which mode are you in, if you're in RAW maybe RAW is displayed with some background color, if dual iso, D.ISO for example, if the bitrate is changes maybe it could say the CBR factor, HDR for cameras that do not currently support dual iso, and so on! Indicators would we great!!! This interface is looking VERY professional!

[stevefal]

Here's an approach that can re-quantize horizontally without looking too busy.

Features:

- arbitrary sub-stop scale - can be 1, 1/2, 1/3 without too much disruption
- noisy tail uses texture instead of height to convey quality
- data can plot horizontally on 1-pixel boundaries if desired  (I prefer quantizing to the tick marks)

Since this design is uniform height, it can also shortened and stacked as an advanced RGBL histobar:

Please note that I have not worked out exact metrics for convenient implementation of these. So tell me if you are interested and I can tweak so layout math is easier.

[Jakobmen]

Here's an approach that can re-quantize horizontally without looking too busy.

Features:

- arbitrary sub-stop scale - can be 1, 1/2, 1/3 without too much disruption
- noisy tail uses texture instead of height to convey quality
- data can plot horizontally on 1-pixel boundaries if desired  (I prefer quantizing to the tick marks)

Since this design is uniform height, it can also shortened and stacked as an advanced RGBL histobar:

Please note that I have not worked out exact metrics for convenient implementation of these. So tell me if you are interested and I can tweak so layout math is easier.

I LIKE..

[Stedda]

I like the advanced RGB a lot. Super quick way of getting the needed info!

[pravdomil]

Good thoughts and what about this?


There will be a histobar with 8 + 2 blocks for MOV recording and 14 + 2 blocks for RAW recording.
It helps you to choose what color depth and exposure you should use.

[stevefal]

Yep, 8<>14 was proposed initially: http://www.magiclantern.fm/forum/index.php?topic=8539.msg79396#msg79396

However I'm sure I'll never ever shoot H.264 again!

[pravdomil]

I'm sometimes confused if the histogram/bar is 14Bit or 8Bit or whatever. There can be some graphical hint, that will tell you what color depth you are using.

btw: one stop of dynamic range is one bit?

thanks for your work

[SDX]

I'm sure I'll never ever shoot H.264 again!

That's great to hear. I, amongst many others, however, will.

[a1ex]

When the lower bins contain nothing but noise, why show them? The histobar shows how much useful range you have with the current settings, and how much you need.

Raw histogram for H.264 is pretty much ill-defined mathematically, because:
- clipping warnings don't match
- the curve is baked in
- I don't understand why it should have 8 stops (especially if the raw one has 5, for example)

Also, the delimitation of stops in the YUV histogram no longer makes sense, because it's totally dependent on the curves (the histogram no longer moves linearly with exposure). This is why I don't print any vertical bars on the YUV histogram.

Besides, I prefer to disable the raw mode when it's not in use (changes to Canon functionality should be minimally invasive IMO - that is, only applied when needed, so if you turn off everything it should behave like standard Canon firmware).

[pravdomil]

- I don't understand why it should have 8 stops (especially if the raw one has 5, for example)

So I would rather show 8 bits instead of 8 stops.
0 block - the first on the left side - shows under exposure - if count of pixels with brightness value 0 > threshold fill this block
1 block - second -  if count of pixels with brightness value 1-31 > threshold fill this block
..
..
8 block - last but one -  if count of pixels with brightness value 223 - 254 > threshold fill this block
9 block - last one - shows overexposure -  if count of pixels with brightness value 255 > threshold fill this block

if RAW is enabled extend this range to 14 + 2 and do the same

[a1ex]

if RAW is enabled extend this range to 14 + 2 and do the same

Why 14 if at least 3 of them are just noise?

[pravdomil]

So only 11 + 2

[stevefal]

@A1ex, any feedback on my last two concepts which were meant to address wants/issues you raised??

I'm still not completely clear about how much we know about the 5th percentile. Here's one with a shaded tail, rather than a noisy tail:

Is the RGBL version interesting? Are you saying that it should not have ruler marks?

[stevefal]

This is my favorite so far. 22-light concept with full-height shaded dark tail:

[wolf]

Looks very good, but what confuses me, is why the blue and red tile runs out of the border.

[stevefal]

Those are intended to represent clipped pixels. In the original concept. I changed the color of the top and bottom tiles, but that gives no indication how much clipping is going on.

In these later concepts, the clip indicators (red and blue) grow outside the bar, at a decreasing rate, to show how much clipping is going on. In a traditional histogram this fact is shown only by the height of a one-pixel vertical line at the edge of the graph - it is very hard, if not impossible to see.

[Audionut]

Hah, are you trolling me?

Of course not!  Just because I have ideas that do not fit in with the "science of GUI design" does not make my opinions nefarious.


I personally think that the EV indicators should remain visible (as in the last screen shot).
The shaded tail looks great.
The RGBL histo is excellent.
The growing over/under exposure indicators are excellent.
I still think 11 is better then 22, regardless of how the indicators should be filled.

[stevefal]

...and 33 would be pretty extreme. I'll try half-filling the lights horizontally to see how that looks.

[Audionut]

Indeed.  I don't think you need to worry about any ticks (ruler marks) to indicate the half stops.  It should be clear when the bar is full or not and what that represents.

I'm still not completely clear about how much we know about the 5th percentile. Here's one with a shaded tail, rather than a noisy tail:

By having the tail shaded, there should only be a lack of understanding of its meaning.  There shouldn't be a misinterpretation.  As long as people aren't misinterpretation its meaning, those that are unsure of it's meaning always have the option of reading the manual.  Blue probably wasn't the right idea as noted by a1ex on the commit.