Red MX vs Alexa vs 16mm Dynamic Range Test

[brawlster]

I don't personally shoot with Red that much (i do do quite a lot of post though), but after a quick search, according to Red's Graeme Nattress:

"We have an exposure mode which shows in red near-clipping data in the raw, and in purple, very dark data in the raw, so you can see at a glance if you are clipping in the raw. This works at ALL isos as it references back to the raw data."

http://reduser.net/forum/showthread....tadata...-quot

...so what he says seems to indicate the false color is indeed at sensor level, not after ISO, Flut etc.

He actually says in your quote "near clipping". Which is not the same as actual clipping. They created a great deal of ongoing confusion when they changed the metering tools with build 30. You'll notice in your linked post, the guy asking the questions is still confused. I went thought the same thing myself. In fact, Graham also clarified in my thread here....

http://reduser.net/forum/showthread....step-backwards

He talks about it only working in "Raw False colour mode" but not in Redcolor.

That thread goes on for 10 pages, with lots of DP's feeling the same way I do.

If you deliberately clip in-camera, you deliberately truncate the dynamic range. To me, this kind of test seems rather meaningless, it just shows how one given exposure setting works on each camera, not what the cameras are capable of. The *get the best out of the cameras for the grade* IS the correct test for the camera's dynamic range.

Is it ? That's your opinion.

I'm often faced with dealing with contrast that is beyond the dynamic range of any camera. Every scene requires an exposure interpretation.

Your argument about deliberately underexposing the RED to get the best out of it in a comparison is meaningless to me in a comparative test. I already know I can trade shadows for headroom. I wanted to compare like for like from the middle of the exposure curve.

Trading shadows for hightlights works only so far. If you have skin tones that are too far down into shadows, it becomes more and more difficult to get them looking right. So again, exposing for highlights is a solution that works for some situations but not all the time and only so far.

In my first setup, without any adjustments other than balancing, the skin tones are exposed the same way in each camera's setup and come out where I expect them to without having to *grade* them up or down. That's what matters to me. If I'd exposed the RED for highlights, I would have had to spend a lot more time in the grade lifting everything up to match.

I chose to test what I wanted to test for. I'm sorry it doesn't tally up with what you're expecting but it tells me what I need to know.

If you really think that RED can be graded and exposed in such a way to be visually indistinguishable from an Alexa, I'd invite you to try it for yourself. And to actually do it yourself rather than relying on other tests out there.

I think you'll find that's not really true where there's high contrast. And your "1 stop" number doesn't really tally with my tests nor many others I think.

Have you seen the SCCE tests ?





John Brawley
Cinematographer
Sydney Australia
www.johnbrawley.com

[Koppel]

As the result of Red religion people only accept one camera as being God.
They just don't believe that some other manufacturer has created a camera with more dynamic range.

[Halsu]

In fact, Graham also clarified in my thread here....

http://reduser.net/forum/showthread....step-backwards

He talks about it only working in "Raw False colour mode" but not in Redcolor.

I've referred to the false color mode all the time, not to redcolor, or to zebra:

"when exposing RED, you can use a false color display (kinda similar to zebra, but more versatile), which shows you what parts of the image are overexposed, and which parts of the image are so low key that noise becomes a problem."
"...so what he says seems to indicate the false color is indeed at sensor level, not after ISO, Flut etc."

...this said, it is indeed silly, if the zebra is post-colorspace, post-ISO, as it seems to be. That's a clear fault indeed.

In my first setup, without any adjustments other than balancing, the skin tones are exposed the same way in each camera's setup and come out where I expect them to without having to *grade* them up or down. That's what matters to me. If I'd exposed the RED for highlights, I would have had to spend a lot more time in the grade lifting everything up to match.

Doing this, you get different results at every ISO setting. In example, if you rate the camera to ISO320, your highlight headroom is two stops less than if you rate the camera to ISO 1280. Again, this doesn't have much to do with what the camera is actually able to produce, as a test like this only tells about highlight headroom at one single arbitarily chosen ISO setting.

I chose to test what I wanted to test for. I'm sorry it doesn't tally up with what you're expecting but it tells me what I need to know.

That's of course perfectly allright, but you shouldn't call it a dynamic range test then...

If you really think that RED can be graded and exposed in such a way to be visually indistinguishable from an Alexa, I'd invite you to try it for yourself. And to actually do it yourself rather than relying on other tests out there.

I'd like to do such a test, trust me, as well as similar tests with a few other cameras. A while ago, we actually discussed how to best make such tests with a local rental facility that has both. We had it all pretty much laid out (real life dynamic range, low light performance, skew/jello, aliasing etc.), but never got to actually commence the tests. Problem is, been busy - and lazy ;-)

...and that's why i was a bit frustrated with your test: it could have been just that, a test whether or not the difference in measured dynamic range actually matters in real life, when exposing for the highlights and pushing the midtones / lows in post to match.

I understand that's not what you aimed for, but i'd still like to see such tests ;-)

I think you'll find that's not really true where there's high contrast. And your "1 stop" number doesn't really tally with my tests nor many others I think.

I based that to a wedge test i saw a while ago, the only one i have seen personally: it showed Red with about 12 stops of DR, Alexa with 13. I think it was in the Red forum. Tried to look for it, couldn't find it unfortunately.

I hadn't seen the test you linked to, which seems to indicate Alexa has 14 stops, giving it one more stop advantage, at least in theory. Would have been nice to see the actual wedge images though, not just bars showing an interpretation of them... (i assume you took the graph from here):
http://provideocoalition.com/index.p...-_scce_charts/

Anyway, the practical usability of the dynamic range isn't that easily interpretable from wedges, it's mainly about noise characteristics in the shadows: how much of that lowest end where the differences are is actually usable in practice with each camera, how much is more or less just unusable noise.

I'd really like to get my hands on such (camera original) footage from both cameras - i guess i'll really need to do the shoot myself to accomplish that ;-)

[Ian-T]

I think all looked good in their own way. Personally...I liked the clean colors and high resolution of the Red MX here. A lot of folks seemed to like Arri the best (it does look good). You can see the strengths of the Arri's latitude especially outdoors on the guy's white shirt. Both 16mm and 35mm seemed noisy. This is also John's (brawlster) work.


http://vimeo.com/24488998

IMPORTANT NOTICE: No Vimeo files are hosted on our forum. By clicking the link below you agree to view content from an external website. We can not be held responsible for the suitability or legality of this material. If the video does not play, wait a minute or try again later.

I agree - Click here to view this VIMEO movie!

[brawlster]

I've referred to the false color mode all the time, not to redcolor, or to zebra:

"when exposing RED, you can use a false color display (kinda similar to zebra, but more versatile), which shows you what parts of the image are overexposed, and which parts of the image are so low key that noise becomes a problem."
"...so what he says seems to indicate the false color is indeed at sensor level, not after ISO, Flut etc."

...this said, it is indeed silly, if the zebra is post-colorspace, post-ISO, as it seems to be. That's a clear fault indeed.

False Colour and Zebra are exposure check tools that are dependent on the *view* that you select within the RED. You can choose to view using "redcolor" or "RAW".

Like Zebra, False Colour is dependent on the view mode and won't necessarily accurately represent sensor clipping.

One of these "views" will give you exposure information that is independent of the ISO and metea data and one won't. RED don't acknowledge what you have described as a fault as being a fault because they argue that you get these tools in RAW.

Of course, they've produced a camera that has a zebra that goes to 109, took away REC 709 monitoring (which used to be a third option in the view menu) and then made it so they have the only camera in the world with a zebra that will NEVER zebra in 100+ range when using it in their substitute view for 709, redcolor. As I say, false color behaves the same way. It's representation is entirely dependent on your monitoring.

Doing this, you get different results at every ISO setting. In example, if you rate the camera to ISO320, your highlight headroom is two stops less than if you rate the camera to ISO 1280.

I don't follow your logic here. The dynamic range of the camera doesn't change with ISO unless you go below the camera's native ISO of 320. It all depends on how you choose to expose the sensor at those ISO's.

If i exposed top down from the highlights at 320 or 800 monitoring in RAW, there wouldn't be any difference in the DR in the final grade all things being equal in post.

This is the real true advantage of shooting RED, that is it most underrated feature. You aren't baking in ISO or white point like you do if you're shooting ProRes on the Alexa. You just need to know how to expose these camera in the first place to get the most out of it.

That's of course perfectly allright, but you shouldn't call it a dynamic range test then...

I didn't realise there was an ISO stand for dynamic range tests. Sorry I don't mean to be antagonistic, but I've done a test that suits my specific needs and parameters, taking into account the post production workflow I have to work with and I'm lucky enough to have a production that let's me share them publicly. I'd be happy for you to show me up by shooting your own tests.

I was looking to see how much range I could get from the cameras in fast-turn around TV, where I rarely get to actually make it to the grade, let alone have time to grade *up* each shot. I wanted to see how they went straight out of the camera before being *saved* by post. Think of it as me exposing for no grade rather than shooting for the grade. It's not always possible to spend the time nursing the grade on every single shot on this production. I would never shoot drama the way I shot these tests. But I am interested to see what the camera can do, not what the camera can do with a lot of time in post tweaking the shots.

I hadn't seen the test you linked to, which seems to indicate Alexa has 14 stops, giving it one more stop advantage, at least in theory. Would have been nice to see the actual wedge images though, not just bars showing an interpretation of them... (i assume you took the graph from here):
http://provideocoalition.com/index.p...-_scce_charts/

I haven't seen any of those actual tests either. That's why I did my own. You can't always trust what you see on the internet either ;-)

Anyway, the practical usability of the dynamic range isn't that easily interpretable from wedges, it's mainly about noise characteristics in the shadows: how much of that lowest end where the differences are is actually usable in practice with each camera, how much is more or less just unusable noise.

Well that's one part of it of what you might want to test. What about.....

What are the impacts on editorial if you expose for the highlights ? How can you ensure an automated rushes transcode produces editorial rushes that are what you intended to shoot ? How can I defeat the curse of LUT's in monitoring on set ? And in post ? How can I expose an image that will give me these best result for the grade when my grade barely allows enough time to do more than a couple of windows in the entire show ?

How nice is the metering in camera ? How well does it represent highlights ? Etc etc.

jb

[antman]

I thought your test was well done, and get what your goal was. The fact you learned what you needed to makes it worthy im my eyes.

If you take three different cars, all stock, add the same amount of fuel to each one, then drive around a track, one is guaranteed to run out first. That in it's self is a test, to see which car gets the best mileage. This is not taking ANYTHING else into consideration. But if you were to open the hood, and start messing with the carbs fuel to air mix, Change out the tires, in order to make your car get better mileage, well than you defeated the original purpose of the first test.

Not sure if the comparison is understandable, but this is how I interpet your test.

On another note, Im seriously blown away by the Alexa, It's the best looking digital footage I have seen up until now. Not saying red is shabby, I personally dont find it as appeasing to my eye as the alexa. Im almost certain that it's the latitude that gives it the edge.

[drapeama]

... Both 16mm and 35mm seemed noisy. This is also John's (brawlster) work.

Wow, downloaded it and watched it 2 times. The comparison makes you see how much the difference there is between several format.
I personally always like to see some 16mm videos...depending on what you're shooting, it can create the mood/atmosphere/ambiance you're after even without additional sets and other stuff. It plays a big role as itself.
Thanks for sharing Ian (and thanks to John to share his work too!)

[Halsu]

Like Zebra, False Colour is dependent on the view mode and won't necessarily accurately represent sensor clipping.

That's not the impression i got from Graeme's answer: he specifically said false color always reflects RAW, and i've always thought this is regardless of the view mode. I don't have Red One here so i could check this right now, but will do the next time i shoot Red.

Of course, they've produced a camera that has a zebra that goes to 109, took away REC 709 monitoring (which used to be a third option in the view menu) and then made it so they have the only camera in the world with a zebra that will NEVER zebra in 100+ range when using it in their substitute view for 709, redcolor. As I say, false color behaves the same way. It's representation is entirely dependent on your monitoring.

This does indeed sound silly, and makes the zebra pretty meaningless if it's in fact true (and i do trust you on this) - but i don't think this is true for false color, so i'll give it a benefit of doubt, at least until i get to test this myself.

I don't follow your logic here. The dynamic range of the camera doesn't change with ISO unless you go below the camera's native ISO of 320. It all depends on how you choose to expose the sensor at those ISO's.

Your logic was: "In my first setup, without any adjustments other than balancing, the skin tones are exposed the same way in each camera's setup and come out where I expect them to without having to *grade* them up or down."

A) You set the camera to ISO 320, expose for the proper skin tones in your monitoring, then use the camera settings in post as is with no further adjustments, you get correct skin tones, but risk blown highlights.

B) You set the camera to ISO 1280, expose for the proper skin tones in your monitoring, then use the camera settings in post as is with no further adjustments, you get correct skin tones, BUT you have 2 more stops of highlight protection (and slightly more noise).

If i exposed top down from the highlights at 320 or 800 monitoring in RAW, there wouldn't be any difference in the DR in the final grade all things being equal in post.

You said you didn't expose for the highlights, but for proper skin tones. In that case, there is indeed difference in highlight protection - the higher the ISO, the more highlight headroom (and less shadow "footroom") you get.

I was looking to see how much range I could get from the cameras in fast-turn around TV, where I rarely get to actually make it to the grade, let alone have time to grade *up* each shot. I wanted to see how they went straight out of the camera before being *saved* by post. Think of it as me exposing for no grade rather than shooting for the grade. It's not always possible to spend the time nursing the grade on every single shot on this production. I would never shoot drama the way I shot these tests. But I am interested to see what the camera can do, not what the camera can do with a lot of time in post tweaking the shots.

I do understand, and as said that's perfectly alright: it's just not a test of the camera's dynamic range capability per se, but rather a test of a set of very specific settings. And as such, i'm sure it works for you excellently, but it's just not at all for i've been hoping to see personally...

I haven't seen any of those actual tests either. That's why I did my own. You can't always trust what you see on the internet either ;-)

Agreed ;-)

Well that's one part of it of what you might want to test. What about.....

What are the impacts on editorial if you expose for the highlights ? How can you ensure an automated rushes transcode produces editorial rushes that are what you intended to shoot ? How can I defeat the curse of LUT's in monitoring on set ? And in post ? How can I expose an image that will give me these best result for the grade when my grade barely allows enough time to do more than a couple of windows in the entire show ?

How nice is the metering in camera ? How well does it represent highlights ? Etc etc.

For the future, one possibility would be to create a custom, highlight preserving curve in Redcine X, and exporting that to the camera. This way, you'd have that "post production highlight save" baked in the camera settings, they should carry over to post in the metadata without further adjustments. It should be WYSIWYG with on-site monitoring, at least in theory. I haven't tested it though.

As far as the other suggestions for testing you have there, excellent ideas. It might be a good idea to record the live output along with the RAW when doing the tests, to get to compare how the feedback in the viewfinder relates to the actual recorded RAW.

Dammit, i guess i'll need to be a bit more active, and indeed shoot those darn tests of mine - it's just a load of work to do so, as you likely very well know ;-)

[Halsu]

If you take three different cars, all stock, add the same amount of fuel to each one, then drive around a track, one is guaranteed to run out first. That in it's self is a test, to see which car gets the best mileage. This is not taking ANYTHING else into consideration. But if you were to open the hood, and start messing with the carbs fuel to air mix, Change out the tires, in order to make your car get better mileage, well than you defeated the original purpose of the first test.

To me, the test is rather as follows: instead of adding the same amount of fuel, you add fuel so that the fuel gauge says you have half a tank. Then you see how long the cars run. It does tell you something, but that something not the actual mileage per gallon / litre...

[Halsu]

This does indeed sound silly, and makes the zebra pretty meaningless if it's in fact true (and i do trust you on this) - but i don't think this is true for false color, so i'll give it a benefit of doubt, at least until i get to test this myself.

Okay, i read through the zebra thread, and everything seems to be pretty logical. The zebra works as it has always worked, in 709 color space, with ire up to 108 (which maintains consistency with i.e. external scopes etc.). Redspace compresses highlights (unlike 709) so that there's never hard clipping unless the sensor itself clips. Redspace always puts these highlight values around ire 98-99, so there's no longer anything that goes beyond ire 100 in the rgb signal (unless R, G, B, gain is used).

Thus, you just need to adjust the zebra to show where you clip - that would be around ire value of 98-99. Set your zebra there (which is the default too), and everything should work as expected. 98 is the new, more or less universal "ISO 320 ire 108", simple as that.

[Halsu]

That's not the impression i got from Graeme's answer: he specifically said false color always reflects RAW

...and by this i mean the exposure check, not the "old style" false color...

[simulacro]

what i can see in the test is that the alexa footage is more similar in dynamic range to the 16 mm footage than the red footage. and based in JB workflow, alexa is closer to cinema in what you see without so much manipulation

[antman]

I may haver read it, but what was the film stock you used on this test?

[Halsu]

what i can see in the test is that the alexa footage is more similar in dynamic range to the 16 mm footage than the red footage. and based in JB workflow, alexa is closer to cinema in what you see without so much manipulation

I find it very likely that simply changing the ISO setting of the camera would have been enough "manipulation" in this case...

[drapeama]

I checked the short film which used to be shot with 6 cameras, and I found interesting to see how dark the "raw footage" could be and how good it looked in the end after post work. I'm surprised that you can have that latitude to start with a dark video and make it look as good as that.
Usually I can do something similar using the HV, but I shoot flat and it's way more bright than this. sure there's a couple of factors like sensor size and all, but you get the point.

[brawlster]

i may haver read it, but what was the film stock you used on this test?

5219 & 7219.


Jb

[1]

I find it very likely that...

The problem, as I see it, with your arguments is that they are based on words & conjecture only,
whereas Brawlster is offering something one can actually look at.

This makes your point a bit mute...

[Halsu]

The problem, as I see it, with your arguments is that they are based on words & conjecture only,
whereas Brawlster is offering something one can actually look at.

This makes your point a bit mute...

I am painfully aware of that.

But i can't help but speak out loud, because the experience i have gathered so far about Red workflow and capabilities makes me rather confident i'm right.

[Gillvane]

I liked the Alexa footage best. IMO, there's no need to shoot 16mm if you can get your hands on an Alexa. If you want that grain, you can just ad it to the Alexa footage in post. I'm pretty sure you could play with the Alexa footage until it looked just like the 16mm footage. But you cannot do the reverse.

If I ever had a budget to shoot a feature (not likely) and was in the position to choose, I'd go with the Alexa.

[1]

But i can't help but speak out loud...

And so you should!

It's good to have these discussions.

[Halsu]

It's good to have these discussions.

Yep, i think so too.

I guess part of the problem lies with the decisions Red made when making their Redspace, which replaced Rec 709. It seems they wanted to make a color profile that gives broadcast legal results in all cases, with no highlights above 100% IRE in the rgb signal, but when doing that, they also made a camera that behaves rather differently than all the other cameras out there - the photographers are used to cameras clipping at IRE 108, not IRE 99, and thus easily overexpose Red.

Also, the ISO setting on Red is different from i.e. still cameras, or from the gain in video cameras: it doesn't actually affect what the camera records, doesn't add analog gain to the sensor. Instead, it just makes your monitoring brighter, and changes the metadata that's read by post apps. As the Mysterium-MX is a rather low noise sensor, the ISO can be upped quite high before noise / grain becomes a problem (if 35mm film-like grain is acceptable, you can lift it way up actually, around ISO 4000 or more). There's a tradeoff in the amount of dynamic range below the middle gray, but in most cases, the results are visually still very acceptable.

So, unlike with the other cameras, when you up the ISO, the result is just that you have more leeway in the highlights, and less in the shadows - more stops above middle gray, less stops below. That's not the same as with most other cameras, where adding gain (or upping ISO) actually changes the exposure, and usually actually results in a reduction in highlight headroom above middle gray.

When one thinks of both of these differences in the camera behavior compared to what people are used to, it's no wonder there's quite a lot of confusion floating around.

I attached a still with various ISO settings, from ISO 320 to ISO 6400... plus one i adjusted with curves to have a decent rolloff of highlights, non-gray blacks as well as enough brightness to see the faces.

In the actual commercial this is from, the color correction was rather dark, somewhere between the ISO 320 and ISO 800 examples.

[Halsu]

Here's one more, at ISO 6400, with medium OLPF compensation (sharpening) and medium noise reduction set in the Red file properties dialog, as well as a regular levels adjustment that simply sets the black and white points to 0 and 255.

Here's all the above at full 1080p resolution, as well as as wmv videos.

http://eki.pp.fi/digivideo/RED/ISO_tests/

The noteworthy thing is, even the most pressed ISO 6400 examples have so low noise that it more or less vanishes altogether when making a decent quality 1080p windows media file, due to compression (some noise was visible in uncompressed preview).

[Halsu]

You'll have to accept that I willingly and knowingly exposed the tests in this way.

I know you did - but if you read the comments people have posted, they now think blown highlights are a feature of Red, not a result of your deliberate decision. I'm not writing to you as much as to the others here, most of which have never used Red or Alexa themselves (or film for that matter).

As far as your post workflow goes, i understand that you want to get the look very close to end result in-camera, and the footage mostly goes through the post using the camera settings, right? That's why i suggested you might wish to try using a higher ISO all the way through (or possibly a custom curve), you'd still have the the WYSIWYG simple production flow, and at the same time, you'd be exposing for highlights.

You refuse to accept my position on the accuracy of Zebras and False colour, and still don't understand why a cinematographer would be upset that 98 is the new 108 (sometimes). You made reference to False colour, not exposure check, which I was the first to mention, as being the more accurate exposure tool.

I'm sorry, that was indeed a mistake on my part, what i meant all along was the exposure check (which is false color too), not the "old" false color. My bad.

And yep, i do indeed understand why you, and many others are upset with clipping at 98-99. I personally do not see that as such a big deal though.

I'd appreciate some professional courtesy instead constantly trying to have the last word and being the most "right" Can't we just disagree on how I conducted these tests and leave it at that ?

We could, if people didn't get a wrong impression, like they definitely seem to have this time. As said, i'm not trying to argue with you, i'm trying to explain my views to the other, less experienced folks here.

[Halsu]

I based that to a wedge test i saw a while ago, the only one i have seen personally: it showed Red with about 12 stops of DR, Alexa with 13. I think it was in the Red forum. Tried to look for it, couldn't find it unfortunately.

I found it again ;-)

It wasn't Red One, but Epic (without HDRx). Here's the chart:



http://reduser.net/forum/showthread....EPIC-and-Alexa

This said, Red One uses the same sensor technology as the Epics as far as i know, both are Mysterium X with similar pixel size. So MX:ed Red One should actually give rather similar results as Epic i guess... or maybe not. My bad for referring to a false memory, sorry ;-)

[Duke]

This said, Red One uses the same sensor technology as the Epics as far as i know, both are Mysterium X with similar pixel size. So MX:ed Red One should actually give rather similar results as Epic i guess... or maybe not. My bad for referring to a false memory, sorry ;-)

The Red MX in the SCCE independent (rather than Red conducted) tests. The F3 was without S-Log.