Arri AMIRA

[sebben]

You mean, it's like an F5 with Alexa color and skin tone that can give you up to 8 stops of overexposure latitude pretty big difference if you ask me.

This is exactly the point. What other camera has so much of its dynamic range attributed to the highlights?

Where I need it most.

 

[cowpunk52]

This is exactly the point. What other camera has so much of its dynamic range attributed to the highlights?

Where I need it most.

Yes, this is a key feature of Arri's cams that people don't often recognize. Looking at total dynamic range definitely doesn't tell the whole story. Digital sensors struggle the hardest to maintain color & detail in overexposure, and Arri stands above the rest in their ability to reach very far into highlights. It's part of the Alexa's "secret sauce," no doubt.

 

[Samuel H]

I understand what you want, but I don't understand why you don't just underexpose to achieve that result. This is not the case if you're recording with a baked-in curve, but on anything recording RAW or log, there is no "above" or "below" DR, it's all the same, you can put the central point of your exposure wherever you want.

Having said that, I still like the images from the alexa and the F35 above most other options.

 

[sebben]

Because then you need to boost the mids, and that makes mucky images. I dont mind trading details in the shadows for nice highlights and clean mids.

I guess I am more interested in stops above middle grey rather than below.

 

[Gillvane]

View attachment 76279
View attachment 76280



Want. Now.





http://www.youtube.com/watch?v=VL1iZgcIaEk
(Features video)

It even has the possibility to use EF mount lenses.

All other issues aside, I find the "pick up>shoot" pitch incredibly appealing. I get the modular design of the Red, customize it, get what you need, replace it, add to it, etc., but I really like this elegant Arri design. I don't want to build a camera with parts, brains and modules. I want to pick it up, and shoot.

 

[Samuel H]

Because then you need to boost the mids, and that makes mucky images. I dont mind trading details in the shadows for nice highlights and clean mids.

I guess I am more interested in stops above middle grey rather than below.

Consider the BMC. It shoots at ISO 200, 400, 800, 1600. Except, it's exactly the same image in all of them (proof here, still not conclusive: http://www.bmcuser.com/showthread.php?5755-ETTR-and-ISO-values-Comparison), the only thing that changes is where you put your middle gray.
ISO 200: 5+ stops above, 7 below
ISO 400: 6+ stops above, 6 below
ISO 800: 7+ stops above, 5 below
ISO 1600: 8+ stops above, 4 below

All you're saying is, you'd rather throw less light at your subject (underexpose) and use a higher ISO (trick your mind into thinking you're not underexposing). Which is fine. And which you can do with any camera shooting log or RAW.

 

[morgan_moore]

This is exactly the point. What other camera has so much of its dynamic range attributed to the highlights?

Where I need it most.

Crazy concept.

Put your mid grey at a lower stop and.. magic.. more room in the highlights with any camera.

To say the massive pixels of the alexa (compared to '4k' cams) hold shadows well to enable a clean grey at a low point on the exposure curve could be an argument that holds water.

The clever thing arri did was make the alexa come out of the box like that creating a mythical filmic rolloff that one has to 'find' in other cams, but is there.

 

[Samuel H]

(exactly; except, if the camera is not shooting log or RAW, the built-in s-curve can screw with you and make this underexposure kind of problematic for grading)

 

[sebben]

I dont want to put mid grey at a lower point. I want mid grey to be mid grey and have more stops of exposure past mid grey.

Crazy concept I know.

 

[cowpunk52]

I don't want to start an argument here, but I'm actually kind of shocked that anyone would think stops above middle grey don't matter and you can just put middle grey wherever and get as much overhead latitude as you want. Sure, you can give yourself extra room in the highlights by under-exposing your mids, but raw or not, if you think that that won't cost you later, then you haven't been at this very long.

 

[combatentropy]

CCD and CMOS sensors do not have curves, as far as I know. They react to light linearly.

You hit a pixel on them with a photon, and they generate a certain voltage. You hit it with two photons: twice that voltage. A thousand photons, a thousand times the voltage. Etc.

There is no such thing as a electronic sensor that "gently rolls off the highlights," like film. It just gets full and clips. To gently roll off, that would mean that the sensor would somehow have to start to rein back the voltage the more photons it gets. Like, at 1,000 photons, it only generates 900 times as much voltage as a single photon. At 10,000 photons, maybe it generates only 7,500 the voltage of a single photon. Etc.

While I'm sure it is possible for some clever engineer to make such an electronic sensor, I doubt that they have. It's much easier to leave the sensor simple and linear and mess with the response later in the pipeline. Maybe Alexa is doing this at the analog-to-digital conversion. But Alexa's sensor is linear, which means it has no highlight roll-off. Which means that if you have a sensor with the same total f-stop range as the Alexa, and record RAW, then you should be able to get the exact same highlight roll-off in post.

Even if you have a sensor with one or two stops less dynamic range, you should be able to approximate the Alexa look. If you have a sensor with one stop less total dynamic range, roll off the shadows a half stop faster and the highlights a half stop faster. If you have a sensor with two stops less total dynamic range, roll of the shadows a full stop faster and the highlights a full stop faster. Or just roll off the shadows two stops faster. You can do it many different ways. It's RAW

 

[starcentral]

I move my middle grey around depending on the situation but I concur with CowPunk it does not come without some associated loss.

When you push an image in post (that is INCREASE EXPOSURE) it is done at the expense of your signal to noise ratio. So those specific tonal values being moved around will not maintain the same level of "noise" or "quality". When you PULL an image in post (that is DECREASE EXPOSURE) you actually improve the signal to noise ratio, but at the expense of latitude.

When you shoot RAW linear gamma keep in mind that everything over 50% signal gets an absurd and exponentially high amount of data allocated to it compared to what you get under 50%. I made this simplified image to show how data is distributed across data bits with a RAW image.

You will notice that there is way more data quantization towards the highlight region, and only a fraction of data allocated towards the shadows and slightly more in the mid-tones. Place your middle grey too low and boost it in post and it will not be very pretty.

This is why they say with RAW to shoot to the right - something you wouldn't do in a million years if you were shooting video non-RAW because once you get too close to your highlight region there is just not a lot of data bits going towards it - unless of course you shoot using some hyper-gammas or even better slog which is a super duper hyper gamma but again nothing close to the power of RAW.

 

[combatentropy]

When you push an image in post (that is INCREASE EXPOSURE) it is done at the expense of your signal to noise ratio. So those specific tonal values being moved around will not maintain the same level of "noise" or "quality". When you PULL an image in post (that is DECREASE EXPOSURE) you actually improve the signal to noise ratio, but at the expense of latitude.

The same thing happens when a camera like the Alexa does it in camera. Noise from the sensor is amplified or muted.

When you shoot RAW linear gamma keep in mind that everything over 50% signal gets an absurd and exponentially high amount of data allocated to it

Which is why I think logarithmic Analog-to-Digital conversion makes more sense. Even 10-bit log seems better than 12-bit linear. Norman Koren, in his article Tonal quality and dynamic range in digital cameras, says that the human eye distinguishes an average of 30 shades per F-stop. 10-bit log, has 1,024 shades since it is 10-bit and distributes those shades evenly across the F-stops since it is logarithmic. So 64 shades per F-stop, twice as much as needed, and room to move them around in post. Bear in mind also that it is the final output image where the shades per F-stop matter. A TV screen usually displays less than 7 stops. So the input F-stops from a real scene with 12 stops of range gets squished into the 5-7 stops of your TV. So you have even more shades per F-stop to play around with.

 

[starcentral]

I wouldn't say its entirely the same to remap values in post than compared to in-camera. Imagine what your software on a PC does when you effectively "re-map" values versus what a camera's internal ASIC might do to process or "clean noise" internally first before storing the image to file? The camera is more than a sensor in a box, there is a serious amount of processing going on in-camera to clean things up.


I've shot a lot using hypergammas on Sony, mostly boosting middle grey in-camera so that in post I can do the opposite and pull it instead. Effectively decreasing the noise as I had described. I shot stuff in 8-bit 420 that had people scratching their heads asking me if it was slog.. nope.


As for shades or data-bits per stop, on a bench it sounds logical to allocate equal amounts to the entire tonal range but if you shot with any cameras other than Alexa and F35 (which have their own highlight secret sauce) you know how hard it can be to get a filmic look from any camera using a curve based gamma. I'm still trying hard to get a "film look" using 10-bit 422 slog2 on the F55 and it is tough, others shooting slog on the F3 would also agree. On occasion when I get to shoot with a RED camera it is almost nearly not as tough to get a filmic look, and I attribute that to the exponential amount of information in the highlights combined with RED's color sciences that pull the image out of the highlights... Same as anyone can do when you shoot a still image in RAW on a DSLR and over-expose it, then use Lightroom or Photoshop to pull down the image. It still looks like something that would have originated on "film".

 

[harddrive]

Which is why I think logarithmic Analog-to-Digital conversion makes more sense. Even 10-bit log seems better than 12-bit linear.

It certainly makes more sense in theory - but is (I believe) far more difficult to do in practice, whereas linear A-D is much more straightforward. Hence, although it might not seem obvious, practically it's easier to engineer linear A-D to 12 bit, then remap digitally to 10 bit log, than go straight from analog to 10 bit log.

 

[combatentropy]

when you shoot a still image in RAW on a DSLR and over-expose it, then use Lightroom or Photoshop to pull down the image

I would do the opposite. Underexpose and pull up.

 

[combatentropy]

It certainly makes more sense in theory - but is (I believe) far more difficult to do in practice, whereas linear A-D is much more straightforward. Hence, although it might not seem obvious, practically it's easier to engineer linear A-D to 12 bit, then remap digitally to 10 bit log, than go straight from analog to 10 bit log.

It's just a matter of numbers, whether 0.123 volts means an RGB value of 652 or 812 (random numbers off the top of my head). I'm sure the devil is the details, as with anything in engineering, but are there not several sensors and cameras that offer 12-bit linear and 10-bit log with the flip of a switch? And Sony's S-Log and Arri's Log C I'm guessing are "loggy" --- mainly logarithmic but with a little S-curvature thrown in to better prep it for film look processing in post.

 

[harddrive]

It's just a matter of numbers, whether 0.123 volts means an RGB value of 652 or 812 (random numbers off the top of my head).

No - because in those cases 0.246 volts (2x) would mean 2x652 or 2x812 respectively, and 0.492volts (4x) would mean values of 4x652 and 4x812 respectively. In either case, you're still talking linear. That's not what you need for log encoding.

(If we assume 0.1 volts means a digital value of 100 units, then for linear a change of 1.0 to 1.1 volt will mean a digital change of 1000 to 1100, 2.0 to 2.1 volts 2000 to 2100 and so on - in each case a change of 100. For log, the change in digital value from 2.0 to 2.1 volts must be less than from 1.0 to 1.1 volts.)

I'm sure the devil is the details, as with anything in engineering, but are there not several sensors and cameras that offer 12-bit linear and 10-bit log with the flip of a switch? And Sony's S-Log and Arri's Log C I'm guessing are "loggy" --- mainly logarithmic but with a little S-curvature thrown in to better prep it for film look processing in post.

But what is the flip of a switch doing? I suspect the analog to digital conversion remains the same whatever, flipping the switch is changing the way it is dealt with later. No different to the gamma circuit in tube cameras, except here the values are adjusted digitally. Input to output value mapping, if you like.

 

[Samuel H]

* As combatentropy said, sensors are linear. There's no mid anything, it goes from nothing but noise to full well in a linear fashion, you can put your subject wherever you want in that scale, the only things that changes is how far you are from the noise bottom, and from the clipped white. So, as starcentral said, there can be a tradeoff. But if DR is big enough (think BMC) you can move that mid point a couple of stops up or down and nothing horrible happens (in fact, as I said before, and as I still have to confirm light-meter-in-hand, that's exactly what you do when you change ISO from 200 to 800).

* Alexa is exactly like that too. And every other camera you can think of. If you want to find something that doesn't fit that description, you have to search pretty deep, and you'll find things like Rambus' binary pixel sensors, which are still pure theory if you ask me (an older version of something offering log readout were the MOSAD designs, multiplexed oversample analog-to-digital conversion, that, AFAIK, went nowhere either).
* The difference with the Alexa is how it tells you to systematically underexpose, and what curve it applies to the data (I haven't measured it, but my guess is that it is log for the most part, then has a rolloff in the highlights).

* Yes, 10-bit log is a lot better than 12-bit-linear, but ADCs are linear, there's no way around that (currently). If a camera is offering 10-bit-log, it's because it converts the original linear data into log at a later stage.
* What you can do is go for higher bit depth. Canon and Nikon both use 14-bit-linear ADCs, which is 4 times as good as 12-bit.

* To me, "expose to the right" means "move your histogram to the right as much as you can, but without having any clipping on the right side". It is what you should always do with a log or RAW system, if you have enough DR in your camera that you can actually avoid clipping. So I would suggest both "underexpose so you have more headroom in the highlights" and "ETTR, that is, overexpose", depending on the scene (former if you have highlights to preserve in your image, latter if not).

* It is entirely the same to remap values in post than compared to in-camera... if you are shooting log or RAW. It is definitely not the same if you are shooting with a built-in s-curve.
* Except, yes, there's a lot more processing involved, and the camera and the PC may do all that NR and sharpening and CA correction and whatnot in different ways. But in principle it's the same (and, if anything, the PC has the potential to do it much better than any camera; that's why RAW developed in ACR looks so much better than internal in every camera that offers both alternatives; terribly slow, of course).

 

[indiawilds]

Surprised to hear that a few people feel it doesn't matter how you expose or where to place the middle gray. Then why have manual exposure mode? A Hero 3 type of auto exposure should be fine?