Reality Check

[Aris_Gavriilidis]

Alright everyone, it's time for a reality check.

Dynamic range has always been a topic of great dispute when it comes to Red's sensors, particularly the latest ones, though I'm mainly focusing on Dragon here. This is a still from an R3D Erich Roland uploaded on Dropbox a while back. It's a Xyla 21 test, Red Weapon Dragon, with Standard OLPF, processed at ISO 800 with DragonColor2 and RedLogFilm. The first three chips of the chart were completely clipped on the waveform in Resolve, leaving a total of 14 stops of usable range the way I look at it. Keep in mind I'm not looking at this with a calibrated monitor, but rather on a 2014 HP laptop, so there may be a little bit more in the shadows there. What is odd about this image however, is that it is identical to the test Phil Holland conducted with the same chart, the same camera and the same settings, except with a different OLPF, his had the STH though it definitely wouldn't make that massive of a difference. On his test however, though it looks identical, the first 3 chips are apparently not clipping, which seemingly results in 17 stops of total range. Now before I go further, I will point out that I absolutely love Phil's tests and his work is even more impressive. His insanely detailed and comprehensive charts on everything from crop factors to lens coverage and ND's have basically taught me the most important aspects of the Red system, they have actually saved my ass on numerous occasions, I've been shooting on an Epic Dragon for two years now and loving it. So I am by no means bashing Phil's test with the Xyla chart, I'm simply searching for a definite and comprehensive result when it comes to the actual range of the sensor. I have never conducted such a test myself, though I would definitely love to, I just don't have access to a Xyla chart, so sadly at the moment I'm limited to looking at other filmmakers' tests and footage. I know the best thing to do is conduct my own tests so I wouldn't be surprised if there's an error on my part when it comes to analyzing footage I haven't shot myself, nonetheless this does look peculiar and Resolve's waveform does tell part of the story. I sincerely hope I don't start a fight, because that is not my intention here. More importantly I also hope I don't offend Phil in any way, his extensive tests have made an immeasurably valuable impact on this community!

Frame from Erich Roland's test I processed, again, according to the Resolve waveform the first three chips are gone.

Here are the scopes I'm getting just as a form of proof

And Phil's test at ISO 800
http://www.artbyphil.com/phfx/cinem...x_REDWeaponTest2015_xyla21_DC2RLF_ISO0800.jpg

I have to say that I am suspecting that Phil exposed literally a hair lower than Erich, thus getting the first chips within exposure in all three channels, but I am still very curious. That's all this is folks, Im posting this as a result of personal curiosity, nothing more, nothing less. I am a Red Dragon guy and I've never found myself in a situation where I needed more range.

 

[Erich Roland]

I'm innocent!
No really... I probably screwed up the test. If those chips are clipped then no conclusions are valid.
Look at Phil's Xyla chart test.

 

[Kenneth C Merrill]

From what I understand (and I could be completely wrong on this) Phil's image of the Xyla with 21 stops is composited by developing the same shot (with the first chip JUST under clip) at different ISO's. That means the bottom several stops are significantly pushed. I guess hypothetically that means the sensor can "see" 21 stops, but in practical terms where you can only develop at one ISO and only push the shadows so far, I get the feeling you top out at about 14/15.

 

[Phil Holland]

Boop.

http://www.reduser.net/forum/showthread.php?154113-RED-8K-Sensor-Test-2017

And more telling:

The two ways I test with the Xyla is first chip full RGB clip and first chip full RGB retained. I've also done first channel clipping, but that's not a real test result though valid with the kelvin of the Xyla.

I actually need to re-test everything all over again now that we have IPP2 "hammered down" and Helium is also rather different than before.

My Xyla's here ready to go, just need to find the day to do everything in a controlled setting.

 

[Anton Shavlik]

I have Eric's R3Ds and in the beta I'm able to recover all but the first chip on his 6k dragon test. Same goes for the helium test.

Eric Roland's tests show 13-14 stops, not including clipped chips, that are distinctly above the noise floor at full resolution. There's visibly signal beyond that, so I suspect shooting 8k and going to 4k will get another stop, and 8k to 2k gets 2 stops, at least in theory. Both the Dragon and Helium sensor are in the 13 stop range, give or take a stop.

Phil's tests show how much signal there is possible to extract using every possible method, averaging 16 frames, noise reduction, downscaling, etc.

Log3G10 should be able to distinguish more latitude than any normal footage would produce, so there isn't a need to develop at different ISOs.

DRAGON:
(same settings on both of these)

HELIUM:

 

[Kenneth C Merrill]

I have Eric's R3Ds and in the beta I'm able to recover all but the first chip on his 6k dragon test. Same goes for the helium test.

Eric Roland's tests show 13-14 stops, not including clipped chips, that are distinctly above the noise floor at full resolution. There's visibly signal beyond that, so I suspect shooting 8k and going to 4k will get another stop, and 8k to 2k gets 2 stops, at least in theory. Both the Dragon and Helium sensor are in the 13 stop range, give or take a stop.

Phil's tests show how much signal there is possible to extract using every possible method, averaging 16 frames, noise reduction, downscaling, etc.

Log3G10 should be able to distinguish more latitude than any normal footage would produce, so there isn't a need to develop at different ISOs.

DRAGON:
(same settings on both of these)

HELIUM:

Thanks for the info. I've always felt like Phil's latitude test was more technical than practical, as it's clear to everyone who shoots with these cameras that they're not getting 7 more stops than any other camera out there.

 

[Phil Holland]

Phil's tests show how much signal there is possible to extract using every possible method, averaging 16 frames, noise reduction, downscaling, etc.

Just to note, in the video for the 2017 test the patches are 100% to scale and published in 8K with no averaging, hell no on noise reduction, etc. One to test what "it can see" and one to test practical usage.

 

[Kenneth C Merrill]

Just to note, in the video for the 2017 test the patches are 100% to scale and published in 8K with no averaging, hell no on noise reduction, etc. One to test what "it can see" and one to test practical usage.

Very cool. To my eye, seems like 15-16 stops.

 

[Aris_Gavriilidis]

Just to note, in the video for the 2017 test the patches are 100% to scale and published in 8K with no averaging, hell no on noise reduction, etc. One to test what "it can see" and one to test practical usage.

Phil could you post a framegrab from the 2017 test video? There's some weird stuff going on at the low end due to Youtube's compression that seems to be eliminating several stops on the low end thus it's not an 100% accurate representation.