Arri Alexa and Mysterium-X...

[Roberto Lequeux]

1 bit does not equal 1 stop. Your bit depth is a range of values available, which fall where they may given the DR of your sensor.

 

[Rob Ruffo]

Oh come on...

You haven't worked with an Alexa yet. Maybe it's overpriced. Maybe it's not. You can't know. And by the tone of your posts you've already made up your mind about how MX and Alexa compare.

No. Not at all. The tone of my posts expresses frustration at the quality of information coming out of Arri, which I find very poor. Its' all about "100 years of history" and "a name you can trust" and so on. Nothing meaningful (except the Prores module) as an advantage over Red+MX, let alone Epic.

The more someone talks about their laurels, the more I feel (and I'm sure I'm not alone in this) that they plan to rest on them. Not good.

I want real info, in part because I WANT them to succeed, not fail. A healthy market category does not have just one player (Red). This is not good for anybody.

If they keep up this strategy only a small segment of much older DPs, who do not feel technically competent to judge individual products on actual merit, and would thus only rely on a "brand they can trust" will chose the Alexa.

Younger guys like many of us here are used to the fact that you can sometimes (within reason) get great deals on gear, and that brand names don't always mean that much. In fact, I'd venture that the younger you are, the more likely you are to inherently distrust most large corporations and their brands. CooLLights, Redrock, Cinevate - they all offer products that compete with much more expensive "trusted established brands" but in some cases offer far superior performance, despite the much lower price. You DO NOT always get what you pay for, sometimes you are only paying for retail mark-up vs. buying direct.

If this happens, and younger DPs working the high end universally reject Alexa, RED will have a monopoly on many market segments. Not good.

 

[Gavin Greenwalt]

1 bit does not equal 1 stop. Your bit depth is a range of values available, which fall where they may given the DR of your sensor.

In linear space 1 bit per stop is the maximum useful range. You can have less but you'll be reducing precision. You can have more but you'll be wasting space storing random noise.

I assume they use 16 bits because they want a nice rolled off highlight. Then you're in non-linear space and need more precision. Hence why they say "We use the full 16 bits of precision." Just not in linear gamma.

Each bit gives you twice as much precision. Each stop gives you twice as much light. If you define the first stop as the measurement between 0 and 1 then you've defined the increment by which all other stops will be measured. Everything smaller than that increment on a linear sensor is just random noise.

But as they say with their dual gain system they aren't striving to have a linear output. They're going after a more film-like, less mathematical aesthetic. With 16 bits they should have more than enough space to add a rolled off highlight and still be able to reconstruct other WBs should they not?

 

[Rob Ruffo]

The exchange rate might tip our decision once our Epic is ready, since we are in Europe. Plus, Arri offers service right around the corner, while I need to go through horrible customs protocols to have a RED sent back and forth…

Good point. An actual, tangible advantage for Europeans.

 

[Rob Ruffo]

Thank you Michael. Actually, thank you very much for your nice
and clear comments.

I am hoping that the Alexa is designed so that even if the current
firmware does bake in the WB, that later firmware builds will not.

Me too. The more that is baked in, the more that can become a mistake, due to fatigue or stress on set, that cannot be fixed later, and the more time it must take before confidently pressing the record button.

 

[Barend Onneweer]

Its' all about "100 years of history" and "a name you can trust" and so on. Nothing meaningful (except the Prores module) as an advantage over Red+MX, let alone Epic.

The more someone talks about their laurels, the more I feel (and I'm sure I'm not alone in this) that they plan to rest on them. Not good.

Since Michael Bravin is the Arri rep posting here, could you point me to the his posts that led you to write the above?

All he says is that they'll give more details at a later stage. Nothing wrong with that. It would be a bad thing to release the camera without giving details. But the camera is not released yet.

Before anything I'll judge a camera by the pictures it produces. Then the workflow and pricepoint. Have a little patience, and you'll be able to do a meaningful comparison.

 

[Rob Ruffo]

1 bit does not equal 1 stop. Your bit depth is a range of values available, which fall where they may given the DR of your sensor.

Exactly. You can represent a 20 stop HDR image using an 8bit jpg, and still be able to see it perfectly well on an 8bit monitor. If you ever saw a web gallery of HDR images, this is exactly what you were doing.

You can map any dynamic range to any bitspace. If a monitor or codec is less than 8 bit, this is where you get problems with subtle gradients, any subtle gradients, even if they represented only a 1-stop difference in the real world from end to end.

Grading is another story - here the more recorded bit-depth you have, the more you can pull things around in post without getting banding.

 

[Rob Ruffo]

Since Michael Bravin is the Arri rep posting here, could you point me to the his posts that led you to write the above?

All he says is that they'll give more details at a later stage. Nothing wrong with that. It would be a bad thing to release the camera without giving details. But the camera is not released yet.

Before anything I'll judge a camera by the pictures it produces. Then the workflow and pricepoint. Have a little patience, and you'll be able to do a meaningful comparison.

Fair enough. I'm just trying to express my vote as to what would constitute a point in the right direction, or at least a direction that would have positive impact on my age group. To my credit, its' not like they've been trying to keep their cam under wraps, and they have been releasing lots of information, albeit not very useful information (in my opinion).

 

[Gavin Greenwalt]

Grading is another story - here the more recorded bit-depth you have, the more you can pull things around in post without getting banding.

Except that in the case of digital capture noise is usually pretty linear through the entire range of the image. So while you could record a 5 stop image to a 32 bit file you are just really accurately measuring noise. The confidence in each pixel is lower than the precision of your measurement.

If you want to manipulate your linear image though and make it non-linear then you need increased precision to avoid rounding errors and be able to reconstruct the original linear image without introducing banding.

Graeme is assuming that the analog gain is being applied linearly and clipping. I'm saying the point of the dual-gain system might be to apply a non-linear curve in order to preserve all the data but bake in rolled off highlights for exposure and WB.

RED's philosophy is to capture linear and process out non-linear. FLUT exposure, FLUT WB etc.
If I'm right then ARRI's philosophy is to bake everything in as non-linear but stores it in enough precision to be able to go back to linear if needed.

Personally RED's philosophy on paper sounds better to me. It requires less storage.

If I'm right about ARRI's system then you can think of it like having a LUT baked into the image, but including the reverse LUT in the metadata. If you had a 32bit EXR and you stored an image of 13 useful stops of information then along with the reverse LUT you could effectively reproduce the original exactly, as long as the LUT didn't clip any information. The bandwidth would be greater but the processing demands would be lower.

This seems to be what Michael Bravin is saying. 'We bake it in, But you can un-bake it. It's non-destructively baked in.'

Here is another analogy:
Log vs Linear.

If you stored a RED image in a 16 bit log file you could reproduce the 12 bit linear image without problem. The LOG gamma would be "baked in". But that wouldn't stop you from undoing the lin->log operation.

 

[Dj Joofa]

No. Not at all. The tone of my posts expresses frustration at the quality of information coming out of Arri, which I find very poor. Its' all about "100 years of history" and "a name you can trust" and so on. Nothing meaningful (except the Prores module) as an advantage over Red+MX, let alone Epic.

The more someone talks about their laurels, the more I feel (and I'm sure I'm not alone in this) that they plan to rest on them. Not good.

Hey Rob, enjoy life, and don't let emotions carry you away. I think both Red and Arri have good specs on their camera. Once both are available you have a lot of time to experiment. Though, Red's track record in doing things in a timely manner is poor, so I don't know how long a wait that will be.

 

[Gunleik Groven]

-removed -
Speculations...

 

[Dj Joofa]

Personally RED's philosophy on paper sounds better to me. It requires less storage.

Hi Gavin. How are you? Are you still the "only one guy"?

I don't know if your above comment is true. Linear data is not the best candidate for compression. That is why JPEG-like stuff have inherently assumed a non-linear mapping to increase data correlation.

Joofa

 

[Barend Onneweer]

Fair enough. I'm just trying to express my vote as to what would constitute a point in the right direction, or at least a direction that would have positive impact on my age group. To my credit, its' not like they've been trying to keep their cam under wraps, and they have been releasing lots of information, albeit not very useful information (in my opinion).

But in the process you seem to be attributing an attitude to Arri (and experienced 'older' DoP's...) of just wanting to sit on their laurels and stick with the things they've done for 100 years.

The Arri D-20 was introduced (as in people put their hands on it...) in 2005. Red made their first announcement (as in people were talking about it...) in 2006. This doesn't say anything about the camera's qualities but it shows at least that Arri hasn't exactly been sitting still hoping to keep selling their legacy film cameras. I understand that you want more information and are dissapointed you're not getting it yet. That makes sense. But all the other sentiments that you're bringing to the discussion really don't make any sense to me.

 

[Gavin Greenwalt]

Hi Gavin. How are you? Are you still the "only one guy"?

I don't know if your above comment is true. Linear data is not the best candidate for compression. That is why JPEG-like stuff have inherently assumed a non-linear mapping to increase data correlation.

Joofa

Yep still just 'that one guy'.

I'm assuming in that statement that Arri is also "Mostly linear". Just with a non-linear white point. To try and keep it an Apples to Apples comparison. I also think it's a safe bet that ARRIRAW is in linearish space since 16bit log would be such overkill.

 

[Roberto Lequeux]

It makes no sense Gavin. Look at the price difference, which allows you to add a RED Rocket. You get instant-dailies, and remain true RAW, save space thanks from REDCODE magic... plus all the other stuff.

This baking and un-baking deal makes no sense, especially when you consider the resolution of Alexa is the delivery resolution so there is no room to throw away rounding errors.

Bigger files, and degrading? Why? I don't see it in the samples.

Arri needs a Graeme. But they don't make enough of those.

 

[Andrew M.]

Having WB completely changeable in post is clearly great, but, even as a dedicated Redite, I have to admit it's only necessary if you really get it wrong while shooting. If Alexa workflow requires a selected white balance that's baked, most shooters who know their light source and how the camera reacts are going to get it close enough to be happily fine tuned in post. And it sounds as if Arri has designed a suite of tools to make that work.

From different comments about the data acquisition on Alexa looks like there are plenty of conversions and adjustments going on in to the ARRIRAW. Probably just to get extra stop and half of DR.

I understand the ability to correct baked data after in post but it all smells like lost of precession or immense amount of calculations involved, thus speed and heat dissipation will be a negative aspect here. Also accuracy and precision of color correction will be very difficult to maintain.

I see picking WB as an initial decision before shooting to be aimed in the shortening the learning curve, but insignificantly.

Same thing in old days in still photography, you had to pick the film stock before hand, so some photographers forced to shoot indoor and outdoor at the same time usually had two cameras armed and ready to go hanging off their necks. These times are long gone.

Arri has a lot of decision to make in short time and the only way out when you push envelope of technology is as usual, the compromise.

Andrew

 

[Scott Crawley]

Me too. The more that is baked in, the more that can become a mistake, due to fatigue or stress on set, that cannot be fixed later, and the more time it must take before confidently pressing the record button.

I really enjoy DI and so I agree that more flexibility is a wonderful thing however there is another nagging voice in my head that grows increasingly shrill. It is related to that last point of yours.

It's always a primary concern of mine that I find a balance between speed and quality. I'm sure there has always been great pressure to move expediently on the prodction day, and I'm equally sure it will always be that way regardless of how quickly we work, so I worry that false confidence will lead to laziness and an overall decline in quality. It is important to take your time.

 

[Gavin Greenwalt]

This baking and un-baking deal makes no sense, especially when you consider the resolution of Alexa is the delivery resolution so there is no room to throw away rounding errors.

You only have to unbake if you screw up the WB. And who knows, maybe the Analog Gain is better than FLUT gain/WB. It's entirely possible. Proof is in the Pudding. Which is why I imagine they're saying "Wait until June" at which point we can see if the extra 33% in bits is worth it.

Keep in mind though that they're recording a 13.5 stop image to 16 bits.

RED is recording a 13.5 stop image to 12 bits.

16 bits has 16x more precision available to it than 12 bit. If they aren't too dramatically altering the image in their WB gain then you've got a lot of room to work with.

 

[Graeme Nattress]

ARRIRAW on the D21 is only 12bits though. Anyone got the specs on ALEXARAW?

Graeme

 

[David Mullen ASC]

All this uproar over WB is a bit much -- in all my years, I've never picked the wrong white balance for a scene, whether I was shooting film or video. The odds are low that you would pick 3200K for a scene and then decide in post that it should have been balanced to 5600K. People who need the options to completely change their mind in post about color don't sound very confident in their choices, or maybe the whole concept of having an artistic concept for how the scene should look -- i.e. previsualization and then execution -- eludes them. Or maybe I'm just old-fashioned and don't get the point of shooting footage in such a vague, non-descript and non-committed approach that it can easily be altered in post in order to "find" a visual concept for the shot after-the-fact. Where's the fun in that?