All the Epic/Dragon Vs. Alexa/Sony threads are belong to us...

[Martin Stevens]

I heard the echos of my name.

Accurately measuring Dynamic Range of camera systems at the level is tricky business and it does need to be measured, not guesstimated.

The briefest way I can explain what the New Calibration and New OLPF provides for Epic Dragon is:

- Approximately 1.3 stops more highlight retention.
- Improved IR Performance
- Far less IR Contamination in the Shadows and Darks, which provides more detail and tonal accuracy down low.
- Eliminates the IR "orbs" around highlights, meaning handles IR reflection and glare better.
- Improves overall color performance across the board, which also means a better straight out of camera image.

Those are the 5 things I can think of. I've had my camera for just a few days now with the new tech and I'll be measuring it soon again. I can safely say that the prototype I tested a couple weeks ago measured a bit over 16 stops of captured and usable Dynamic Range in places you'd "expect it". There's actually even more than that that you can measure, but you wouldn't shoot flat images that way.

That's where it stands right now. The color is now just fantastic. It was good before and now it's even better.

I thought the new Dragon OLPF also (mostly) eliminated the grid of pink dots when the a lens is pointed at the sun etc. at say f22.

 

[Noel R.]

Since comparisons to Alexa are inevitable, and Arri themselves dismiss pretty much every other test people use to measure the dynamic range of digital cameras, it might be informative to see Dragon tested using this contraption.

Zacuto's "The Great Camera Shootout 2011" used it to measure the cameras involved, including the R1MX, but that's the only time I know of it being used to test a camera from RED.

I think it would be a fair comparison to use this test, not necessarily because it is "The" standard, but a standard we can all relate to, especially considering many of us are using the Alexa as a basis for comparison.

I think once the dust settles from all of this what we're going to see is that the Dragon with the updated OLPF and calibration comes in at a measured 16.5 stops of usable dynamic range.

The Alexa will measure at about the same end to end but the reason that Arri doesn't publish this as the spec and instead says 14.5 stops is they know the stops at the top are purely rolloff (low con?) and not capable of recording usable detail and the bottom stop is too noisy, we already know the Dragon is about a stop cleaner in the shadows so this makes total sense. So although on a chart you may be able to count the same number of patches on each camera you'll be able to actually use and have access to those extra 2 stops on the Dragon in a way that the Alexa doesn't. It will also be interesting to try some low con and light diffusion to see what affect this has on Dragon highlight rolloff.

I'm also curious to see how low-con affects the apparent dynamic range in the image. I think I get what you mean by usable dynamic range, It's not something that should be dismissed, but it does add to the confusion.

Still dragon shines at 5000ASA... alexa dont...

https://vimeo.com/87794102

So to me you have to take resolution into account when you talk range. Dragon downscaled to 2k is slick at far higher iso's than alexa files at the same iso and 2k due to the huge downsampling going from 6k to 2k. So usable range is always something that has to be discussed together with resolution as noise goes up and down with image size.

Edit one more with lower iso... low like 2500 and differnet white balance... whenever 2500 ISO that became a low I do not know...


https://vimeo.com/87804971


This camera makes the street outside my office look like a backlot in hollywood..

Nice test Bjorn, I'm impressed with what I see at 5000 ASA. I would call that usable. At the end of the day, I agree with what has been said, Dragon has a lot going for it other than dynamic range.

 

[Nick Morrison]

Bjorn those tests are gorgeous. Looks like a Fincher movie! Agreed, scaling down that much DR from 6K to 2K is clearly a massive plus for Dragon, and clearly leads to very clean usable results even at high isos. Your footage is such a nice reminder of this. Thanks for sharing.

 

[Nick Morrison]

Nice test Bjorn, I'm impressed with what I see at 5000 ASA. I would call that usable. At the end of the day, I agree with what has been said, Dragon has a lot going for it other than dynamic range.

Agreed. I think the "clean" look & lack of noise, resolution, color, and "elasticity" in the image are all major pluses for Dragon, beyond whatever DR it has.

Curious about what lowcon could do to Dragon imagery and DR as well....surprised it hasn't come up yet!

 

[Blair S. Paulsen]

Sensor DR vs system DR

Sensor DR vs system DR

Keep in mind the optimization of the Dragon/new OLPF combo is just beginning. The part of the signal chain where the bayer pattern data is turned into an RGB image by the algorithm is a key determinant of signal to noise ratio in the RGB image you are measuring.

Unless you are using a custom debayer of full raster ARRIRAW, the Alexa images you are evaluating have already been down sampled and made into RGB per internal processing crafted by ARRI to create "pleasing" pictures. This likely includes highlight compression, noise reduction, colorimetry offsets, etc (all carefully engineered to mimic Kodak 5219 spectral and density response characteristics).

RED pretty much just turns sensor output into a data file and writes it to storage. How the RGB image derived from that data looks on the monitor or what code values are written into an RGB format like DPX or ProRes is yet to be determined. This RAW development phase, starting with the decoding algorithm, is what determines how many stops one can measure in the derived RGB image one is using for comparison. ARRI has certainly improved their IQ over the last 4 years by optimizing the RAW>RGB processing metrics - and that's not just my opinion, it's what most colorists report. Assuming the newer tech in the Dragon sensor has better potential DR (highly likely), I'd expect the DR in the RGB derived image to continue to improve steadily over the next couple of years.

BTW, I really believe that in motion digital imaging applications using existing tech, signal to noise ratio is the more illuminating measuring stick - not the stops of DR found downstream of a selected processing path.

Cheers - #19

 

[Nick Morrison]

BTW, I really believe that in motion digital imaging applications using existing tech, signal to noise ratio is the more illuminating measuring stick - not the stops of DR found downstream of a selected processing path.

Cheers - #19

Love your posts, Blair. Thanks for sharing.

 

[jake blackstone]

Unless you are using a custom debayer of full raster ARRIRAW, the Alexa images you are evaluating have already been down sampled and made into RGB per internal processing crafted by ARRI to create "pleasing" pictures. This likely includes highlight compression, noise reduction, colorimetry offsets, etc (all carefully engineered to mimic Kodak 5219 spectral and density response characteristics).

RED pretty much just turns sensor output into a data file and writes it to storage.

ARRIRAW by definition is just that-RAW, just like R3D. The main difference being, the ARRIRAW is uncompressed, where R3D is compressed. Otherwise, they are pretty much identical in the way, that they are both recordings of RAW sensor data. I have no idea why there is a need to constantly misrepresent ARRI camera technology.
And if you are going to compare the two cameras, then compare R3D with ARRIRAW and not Prores.

 

[Nick Morrison]

ARRIRAW by definition is just that-RAW, just like R3D. The main difference being, the ARRIRAW is uncompressed, where R3D is compressed. Otherwise, they are pretty much identical in the way, that they are both recordings of RAW censor data. I have no idea why there is a need to constantly misrepresent ARRI camera technology...

Jake think what Blair means is how this RAW data is processed in camera as it's recording...ie that what's always evolving/being tweaked, whether it be at Arri or RED. Notice how far RED has advanced w/ processing Dragon footage in only last 6 months. The latest firmware build is by all accounts an important step forward.

 

[David Battistella]

. It will also be interesting to try some low con and light diffusion to see what affect this has on Dragon highlight rolloff.

Im curious to see this as well Evin. 1/8 dig icon would be a nice start, but the way the highlights are rolling off with the new OLPF right now, I'm thinking there is much less need for even the mildest diffusion.

As far as the "stop talk" goes.

I'm im seeing phil's tests and John Marchants 21 step wedge chart tests really showing all the DR we need. There is going to be some tweaking in terms of where we place middle grey in that range. In terms of usable DR, obviously we want as much as we can get, but most of what we actually see in films is far less than the usable range once graded.

16 stops in some scenes will just look unnatural, like IMHO so much of that bad HDR photography in the past few years.

Battistella

 

[Evin Grant]

My favorite filter for this is the 1/16 HD Classic soft, it's so subtle and adds just the right amount of bloom to practicals.

 

[David Battistella]

My favorite filter for this is the 1/16 HD Classic soft, it's so subtle and adds just the right amount of bloom to practicals.

Thanks Evin, How does it handle hard backlight? I've seen some filters just wash out in those "contro luce" situations.

 

[jake blackstone]

Jake think what Blair means is how this RAW data is processed in camera as it's recording...ie that what's always evolving/being tweaked, whether it be at Arri or RED. Notice how far RED has advanced w/ processing Dragon footage in only last 6 months. The latest firmware build is by all accounts an important step forward.

That is the whole point of RAW recording, be it ARRIRAW or R3D. By the very definition, RAW recording is just a recording of what sensor sees with no processing of any kind done to it. That is the reason why RAW can be re-processed i.e. "developed" over and over using new algorithms i.e. "new color science". If any type of pre-processing is applied, then it's not RAW any longer by definition. In that case the way the image looks becomes "baked in", like the Alexa Prores recording

 

[David Battistella]

Nice stuff Bjorn. How much did it cost you to have the water truck go by? LOL in other words, you picked the right night....

Battistella

Still dragon shines at 5000ASA... alexa dont...https://vimeo.com/87794102So to me you have to take resolution into account when you talk range. Dragon downscaled to 2k is slick at far higher iso's than alexa files at the same iso and 2k due to the huge downsampling going from 6k to 2k. So usable range is always something that has to be discussed together with resolution as noise goes up and down with image size.Edit one more with lower iso... low like 2500 and differnet white balance... whenever 2500 ISO that became a low I do not know...https://vimeo.com/87804971This camera makes the street outside my office look like a backlot in hollywood..

 

[luigivaltulini]

brief summary .. new - old ..
same curve applied in the x-REDCINE for 2 shots .. be interesting to see how they respond calibrations ..

[/URL][/IMG]

 

[Hrvoje Simic]

Interesting. Same comparison with MX would be very informative.

 

[David Battistella]

Luigi,

Did you have the denoise on in RCX? Old seems less noisy in the darks for sure. I guess the green is coming from the green OLPF. Your 49 and 50 clips are interesting too.



Battsitella

 

[luigivaltulini]

Luigi,

Did you have the denoise on in RCX? Old seems less noisy in the darks for sure. I guess the green is coming from the green OLPF. Your 49 and 50 clips are interesting too.



Battsitella

David, no nothing applied to denoise and no white balance ..
It in fact OLPF old is less noisy ..
I still had on Carbon 5.1.35, while the new have the 5.1.38 ..
Should I do some tests with the same firmware, and maybe leave the light to maximum and use the MM with ND ..
I would like the new has less noise in the blacks as the old ...

luigi

 

[luigivaltulini]

Interesting. Same comparison with MX would be very informative.

This is using the Motion Mount with 8ND. (2.4) and t1.8 wide open
I put a Scarlet for comparison.
No white balance did ..
These will be used when it comes ADD, to see what can be done.
you can play with the raw below.

https://copy.com/fgwGUUKsM9gE

[/URL][/IMG]

 

[David Battistella]

Caspita!

Which dragon is old and NEW? E1 is new? (don't tell me that) E2 is old? (prefer) I hope it's the other way around.

Even the scarlet can be pulled back to something reasonable.

Maybe we need the new version of RCX? I'm on Beta 23

 

[luigivaltulini]

caspita si..

A is old David and E in new


just for fun Dragon and scarlet :

https://copy.com/p82Gg3MmKNDx

[/URL][/IMG]