DRAGON - ITS THE REAL DEAL !!!!

[Mark Toia]

Mark, did you discover IR polution over Dragon sensor ? I notice that you shoot on pretty high contrast and in the forest and you dont use any NDIR filters.

Rivai

Nope, I used no IR protection, Opposite actually, I was looking for the IR pollution, I wanted it to come out and play.
It did show up, but I was cranking the VND to the max to see it. Its a lot better than the MX in this area, and told by Jarred that its going to be sorted when released.

 

[RivaiC]

Nope, I used no IR protection, Opposite actually, I was looking for the IR pollution, I wanted it to come out and play.
It did show up, but I was cranking the VND to the max to see it. Its a lot better than the MX in this area, and told by Jarred that its going to be sorted when released.

Good to know that. Thanks for your hard work Mark ! Appreciate it very much

Rivai

 

[Robert Ruffo New]

Hi Mark! Thanks on behalf fo all for doing this. When you say cranking the VND to the Max, how many stops is that? Like a 1.8? A 2.1?

 

[Alexandros Tsoupras]

Read-reset of the Dragon sensor is half that of the MX sensor... so skew will be half as much.

Jim

Thank you Jim. Thank you Jarred. Thanks to all the RED team. Thank you Mark for the test.
Jim what about IR pollution improvements?

 

[Robert Ruffo New]

You are so right, Chris.
The limiting factor is actually the displays we use.

Well no, it's not. You can grade footage with power windows, and have different F stops in different regions in the final grade. Power windows aside even a very poor contrast monitor can still display any number of f-stops of latitude. A low contrast display will not clip your whites or crush your shadows into noise, as will a low dynamic range capture device, it's not all the same thing. A low contrast monitor will only look washed-out, as it remaps incoming images to it's available range of brightnesses. You could see information from an image with 22 stops of latitude displayed on a monitor that has only one f-stop of contrast. Maybe it would look like crap, but you would still be able to see information present in the highlights and shadows as captured. If your capture device clipped at either end, then that information would not be there.

The real issue is how many stops can you create in a grade and still have an image that looks pleasing - not washed out - when displayed on most consumer devices. That varies based on material, but without using power windows I'd say maybe 11 at the most, although they are getting better.

The rest of what Chris says is very useful.

 

[Mark Toia]

Hi Mark! Thanks on behalf fo all for doing this. When you say cranking the VND to the Max, how many stops is that? Like a 1.8? A 2.1?

7 stops. whats thats equal too... Say a ND 9 and a 1.2 + 7 stops? yes. no?

 

[Robert Ruffo New]

I think what a lot of people here overlook is the fact that the EPIC Dragon can shoot so many pixels per second.
This is not trivial, folks!

Sure, there are high-speed cameras out there (Phantom comes to mind) that can shoot over 1000fps...but for how long?
Sure, there are pro Nikon and Canon cameras that can shoot over 20MP full RAW files, but at what speed...and maintain that speed for how long?

Consider the heat build-up, the limitation of the recording media's speed, the processor, the buffers...
RED has managed to out-perform everyone else, and bring a product to the market that is an order of magnitude more capable than the competition.

By the way, there are 3 reasons that I can think of as to why the Dragon's RedCode is so highly compressed when shooting high frame-rates:
1. The recording media (Red Mag SSD) and the wires that connect it to the camera are simply not fast enough to record more pixels/second. Not even the 512GB mag.
2. Even if the mags could handle less compressed footage (5:1 or better, at 60fps or higher, 16MP or higher), they would fill up so quickly that recording times would be down to few minutes.
3. You'll barely see the difference between 17:1 and 8:1. You'll barely see the difference between 8:1 and 3:1. Why bother with extra terabytes of footage when the gain is almost none-existent?
I'm relying on Mark's notes and sample footage for this one.

Kudos to the RED team, engineers, management and marketing.

And as a bonus, besides getting a camera than can push out pixels like no other - it also gives you better DR and cleaner high-ISO shots.
And the camera size stays the same. And the upgrade costs is VERY reasonable.
Truly amazing feat.

Compression noise shows up more when you do certain kinds of color grades, which Mark did not do as they are not possible in RCX - but sensor noise makes the compression "break" sooner, so the Dragon, with its lower noise, will compress better.

 

[Eric Z]

The real issue is how many stops can you create in a grade and still have an image that looks pleasing - not washed out - when displayed on most consumer devices.

Robert,
That's what I meant to say. Displays need to get better - real 10bit color, higher color gamuts, etc.

 

[Robert Ruffo New]

7 stops. whats thats equal too... Say a ND 9 and a 1.2 + 7 stops? yes. no?

7 stops = ND 2.1 - that's the highest we have ever use - thats a lot.

MX Epic starts showing IR issues sometimes at ND 0.9 (3 stops) and almost always at 1.2 (4 stops). That's a crazy-huge improvement.

 

[Robert Ruffo New]

Robert,
That's what I meant to say. Displays need to get better - real 10bit color, higher color gamuts, etc.

10 bit has nothing to do with dynamic range. a TRUE full 8 bit display (many, even most, are actually less than that) will have a very hard time showing perceptible banding, ever. A 10 bit display will not make lower contrast images (images that contain information from a very wide range of captured light-levels, where nothing was crushed-out at either end) look any better, nor will higher gamut. The benefits of 10 bit and wide-gamit are completely unrelated, especially at a perceptual level.

 

[Antony Newman]

2 x Sensor Read rate + DSMC FAN 2.0 + 17:1 = ~600 fps @ 2K in 2014?

AJ

Read-reset of the Dragon sensor is half that of the MX sensor... so skew will be half as much.

Jim

 

[Mark Toia]

7 stops = ND 2.1 - that's the highest we have ever use - thats a lot.

MX Epic starts showing IR issues sometimes at ND 0.9 (3 stops) and almost always at 1.2 (4 stops). That's a crazy-huge improvement.

sorry, i may have lead you astray, I meant thats what we were pushing too.. 7 stops ND. Thats when it got to the point were I didn't accept the pictures I was getting.
We were noticing IR starting to come in around 5 stops... not as bad as MX.. but it was still there, Anyways, i don't think the shipped cameras will have the issue. IR will be sorted by then I sure.

 

[Brant Hadfield]

I will try this question again. Nobody's talking much about color or aesthetic. Was the color science used to decode this footage of Mark's specific to the Dragon sensor, or was it the same RG3 we all know and love? And is there any indication of what stage of development the Dragon color science is?

 

[António Fagundes]

I will try this question again. Nobody's talking much about color or aesthetic. Was the color science used to decode this footage of Mark's specific to the Dragon sensor, or was it the same RG3 we all know and love? And is there any indication of what stage of development the Dragon color science is?

Mark said in his OP that this test was mainly focused on DR and low light.

 

[Andy Roberts]

I will try this question again. Nobody's talking much about color or aesthetic. Was the color science used to decode this footage of Mark's specific to the Dragon sensor, or was it the same RG3 we all know and love? And is there any indication of what stage of development the Dragon color science is?

From Jim...

It has been stated a few times already but I'll continue to remind everyone that these are 1st tests... and we are really proud of the results. But fear not... we have a thousand knobs to turn to extract more from this package.

Somewhere in the dozens and dozens of pages of this and the other thread, it SEEMS that Mark was using RedColor3 and RedLogFilm.

 

[egon touste]

great shots..the camera is real!! not to nit pick, but the shot that you label "underexposed and pushed on purpose.. why shoot at 250iso ..this (meta data) will make the image look darker on lcd while shooting, and lean you towards over exposing.. ..exposing for 6400iso on lcd would be pushing ..or how are you rating ISO for this? (or is it a typo etc)

 

[steve green]

Mark,

What was the reason you did many of the night shots at 250 iso and the pushed to 4000 iso in RCX? Is there a difference in just shooting the scene at 4000 to begin with? I've been in situations that there would be a lot of nervous agency folks on set asking a lot of questions if I were to shoot a scene so underexposed.

Also, can you give us the max frame rates at the various resolutions?

Thanks again for the truly fantastic footage.

Steve

 

[Brian Boyer]

I will try this question again. Nobody's talking much about color or aesthetic. Was the color science used to decode this footage of Mark's specific to the Dragon sensor, or was it the same RG3 we all know and love? And is there any indication of what stage of development the Dragon color science is?

Mark said in his OP that this test was mainly focused on DR and low light.

I think Brant's question is what exactly are we looking at?

It's my understanding that Dragon has a new color filter array. Not just new in the sense of being bigger with more pixels but also new in formulation of the individual red, blue and green areas.

The shots Mark (lightly) graded starting from Redlogfilm would benefit from having all the gamma info available in the image and would certainly be getting improved color from the CFA but may still be held back by an unoptimized RC3.

As for the shots that started from RC3/RG3, several possibilities come to mind.

1) The old color science, RC3/RG3, as seen through the new, improved CFA. This would presumably produce a different picture from Dragon than RC3/RG3 through the old MX CFA and could possibly yield better looking color despite being the older color science.

2) New color science, as seen through the new CFA but tweaked to look like RC3/RG3 through the MX CFA. This would presumably be indistinguishable from MX and done to ensure consistency between Dragon and MX at the same settings.

3) One of the first two scenarios is correct but a forthcoming (RC4/RG4?) tuned specifically to leverage the new CFA will yield even better out-of-camera and graded results. I think this is what many people are waiting for and were hoping to see in Mark's footage. This has nothing to do with what he tested and everything to do with what was available at the time.

Jim said they're still tweaking, as they always continue to do. It would be interesting to know just how Dragon is interpreting RC3/RG3 through the new CFA.

 

[Mark Wilkinson]

I think everyone should stop and think about the numbers that we throw around routinely these days.... 14 stops... 16 stops, 18 stops of dynamic range... Trying to even create a 16-stop lighting condition is extremely difficult.

Let's be pragmatic and just look at the numbers of what we are saying...

If we start with metering the darkest something in frame at an f1.0, we'd need to have something bright come in at an f256 to be 16 stops brighter! That f number doesn't even compute to our photographic brains!

Here are 20 f-stop numbers listed out... So, from an f.35 to a f256 represents a 19-stop range!

.35, .5, .7, 1.0, 1.4, 2.0, 2.8, 4, 5.6, 8, 11, 16, 22, 32, 45, 64, 90, 128, 180, 256

Who can even contemplate an f256! So sliding that first scenario away from those ridiculously high f numbers, let's shift our low light reading to an f.35... Okay, I actually have light meters that go that low... But even then, that still would put a highlight source in shot at an f90...

That is some massive contrast ratio to try to capture!

Finally, anyone even want to venture the largest dynamic range the best display mediums, be it film or electronic, can reproduce? A hint, it's no where near 16 stops!

Now I will play devil's advocate here and encourage anyone with a spot meter to go out and find within 1 visible "scene" a measurable 16 stop "shot"... I'm interested to hear the results! It may even open a few eyes!

Fair enough, but I think the 16 stops might not go as far as one might hope, either.

If one were shooting an interior and wanted to be at 2.8 for key... but there is a window in frame out of which we see the bright, cloudy sky. f45 would be a pretty common meter reading for that. Which would be at the very top of the 8 stops. In those clouds would be detail that would probably be gone since we'd be at the very end of our 16 stop range.

You could sacrifice your shadow detail and shift your key lower in your 16 stop range. But what if you're shooting a commercial for a black and gray velvet quilt and the client needs to see all the detail in the fine Dutch weave?

This exact thing happened to me last week on a shoot.

Ok, no it didn't.

 

[António Fagundes]

Fair enough, but I think the 16 stops might not go as far as one might hope, either.

If one were shooting an interior and wanted to be at 2.8 for key... but there is a window in frame out of which we see the bright, cloudy sky. f45 would be a pretty common meter reading for that. Which would be at the very top of the 8 stops. In those clouds would be detail that would probably be gone since we'd be at the very end of our 16 stop range.

You could sacrifice your shadow detail and shift your key lower in your 16 stop range. But what if you're shooting a commercial for a black and gray velvet quilt and the client needs to see all the detail in the fine Dutch weave?

This exact thing happened to me last week on a shoot.

Ok, no it didn't.

If you want detail in the clouds, then that's where you start. And that is why you would ND the window or light the subject to pull your 2.8 key to whatever contrast ratio falls within your 16 stop range of the Dragon, the highest being the clouds, then work your way back... you would probably want to read your subject at f5.6 or f8 (if the clouds are at f45) , so that your lower dark parts of the frame wouldn't fall too much under that 16 stop ratio. IF you want to see EVERYTHING in the frame in such high contrast scene (which I think is a bit boring cinematography, but that's just me...)

Nothing complicated when you have 16 stops to work with.

The MX had already plenty of DR to work with, Red decided to pump up some more, fine. But there are more important things that we need to see from this sensor.