NOISE: Scarlet Dragon versus Scarlet-W

[Stephen Pruitt]

Hey guys. . .

We currently own a Scarlet Dragon and love it.

I'm puzzled because the stated specifications of the two cameras, outside of data-rate, are essentially identical. Same dynamic range. Same signal-to-noise ratio. Given that sameness, how can one state that the Scarlet-W has less noise than the Scarlet Dragon? I mean, if it had lower noise levels, the signal-to-noise ratios would reflect that, right?

Thanks.

Stephen

 

[Hugh Scully]

The story is that the new processing on the Scarlet-W improves the noise levels. I have not seen any comparisons and wonder if it's not the OLPF. I have a Scarlet Dragon too (love it) and shot at 400 ISO to keep the shadows noise free with the STH OLPF. After changing to the Standard I can shoot at 800 with the same clean shadows. I wonder if that doesn't have a lot to do with it.

 

[Phil Holland]

Better hardware, more optimized optical path. It's not the same camera really, though it is the same sensor tech. And data rates do play a roll as well.

 

[Hugh Scully]

Phil, could say more about the optimized optical path? How does it differ from SD?

 

[Stephen Pruitt]

But, Hugh/Phil, if the two cameras' measured noise and range specifications are EXACTLY the same, HOW ON EARTH COULD THEY REALLY BE DIFFERENT with respect to range and noise?

Granted, I'm "just" a scientist, but honestly that makes absolutely no sense to me. Clearly, it would be to RED's great advantage to advertise the better specifications (if they, indeed, were better). Since they don't, I'm forced to the conclusion that they really do have the same range and noise characteristics.

Now, would I prefer the higher data rates and smaller size? Absolutely. I just can't believe that on the two characteristics of a camera that really matter to me (noise and range), SW is better than SD. Both are touted as having identical specs in those two areas.

Stephen

 

[Tim Daust]

seeing is believing...

 

[Hugh Scully]

Where can it be seen? I think Stephen has a compelling perspective. Any side by sides out there? I've heard about improvements but only anecdotally as relative to general DSMC2 improvement.

 

[Sabovic Adis]

But, Hugh/Phil, if the two cameras' measured noise and range specifications are EXACTLY the same, HOW ON EARTH COULD THEY REALLY BE DIFFERENT with respect to range and noise?

Granted, I'm "just" a scientist, but honestly that makes absolutely no sense to me. Clearly, it would be to RED's great advantage to advertise the better specifications (if they, indeed, were better). Since they don't, I'm forced to the conclusion that they really do have the same range and noise characteristics.

Now, would I prefer the higher data rates and smaller size? Absolutely. I just can't believe that on the two characteristics of a camera that really matter to me (noise and range), SW is better than SD. Both are touted as having identical specs in those two areas.

Stephen

Maybe it's only valid for same resolutuion.

 

[Michael Tiemann]

Hey guys. . .

We currently own a Scarlet Dragon and love it.

I'm puzzled because the stated specifications of the two cameras, outside of data-rate, are essentially identical. Same dynamic range. Same signal-to-noise ratio. Given that sameness, how can one state that the Scarlet-W has less noise than the Scarlet Dragon? I mean, if it had lower noise levels, the signal-to-noise ratios would reflect that, right?

Thanks.

Stephen

I just a lowly REDUSER, but I owned an EPIC DRAGON and a SCARLET DRAGON before upgrading to two Weapon MGs and buying a Weapon W for good measure. The image quality of DSMC2 DRAGON vs. DSMC DRAGON is instantly noticeable. I think the path that I took explains some of the science along the way. I started with what would later be called the LLO OLPF. I liked the images, but to love them I had to give in to accepting certain "aesthetics". The colors were better than anything I had, but they were still stuck in the "uncanny valley" of realism for me. I learned (as many on this list did) that part of our problem was that we were underexposing our images. The recommendation that we shoot somewhere between ISO 800 and ISO 2000 as one large equivalency was really false. Moreover, there was a considerable amount of shadow noise that was really quite colorful--bright red pixels and bright blue pixels where really there should have been at most dark red and dark blue, though really dark grey. Lots of folks noticed the same, and we got two things from RED: Dragon Enhanced Black--a post-processing step that substantially changed the "colored popcorn" in our shadow detail, and a Skin Tone Highlight OLPF that provided much better color rendition--along with the caveat that we should rate it at ISO 320. And more and more folks began to agree that LLO was really better at ISO 800 than ISO 2000. (High-quality noise reduction software could tame ISO 3200 and beyond, but if we're talking about one-light simple digital film development, ISO 320 for STH and ISO 800 for LLO represented much more responsible settings than ISO 2000.)

When I went to DSMC2 (with STD OLPF), I immediately noticed I got color that was > 90% of the DSMC STH OLPF, and I could rate that at ISO 800 without trouble. There was also much, much less "colored popcorn" in the image. I suspect that this may in part be due to the fact that when the sensor is throwing lots of random red and blue pixels into the mix, the wavelet compression does its best to capture that as "detail", which means that it's also taking away from bandwidth that could be used to store actual image detail. A cleaner readout of the sensor leads to a much higher fidelity image for the same compression rate. And a cleaner readout of the sensor also means less actual noise in the image to begin with. Add to that 2x the possible data rate (140 MB/sec vs. 70 MB/sec) and you really are talking about two different cameras.

I will say that one person's noise is another person's texture. I do think that DRAGON's noise is very organic, and can be a good look (if all other colors are behaving themselves). The latest RWG/LG10 color stuff has completely solved the color problems for me of DC2/RG4. I'm very happy about that.

I will also say that a reason you probably don't see much of a spec change is this: the noise that knocks out the blackest chip of a test chart is due much more to math noise (which cannot be solved) than the colorful noise introduced by the readout of less-than-black chips. DR specs really only measure what the number of chips you can read (yes or no) rather than what's the level of noise in the bottom three chips you can read. If there were such a spec, DSMC2 would be clearly superior to DSMC. But DEB, lower ISO ratings, noise reduction, etc., can produce very clean images on DSMC, which we have seen. DSMC2 just makes it that much easier, at higher ISOs, which is nice.

 

[Stacey Spears]

But, Hugh/Phil, if the two cameras' measured noise and range specifications are EXACTLY the same, HOW ON EARTH COULD THEY REALLY BE DIFFERENT with respect to range and noise?

Specs are normally based on theoretical numbers from parts. e.g. Its common to use the S/N ratio from a DAC (Digital to Analog converter) when talking about audio devices. The 16+ stops is most likely measured in raw space.

Best thing you can do is rent a Scarlet-W and test for yourself.

I can tell you that in my testing, using a DSC Labs Xyla chart, that my Weapon Dragon was cleaner than my EPIC Dragon. This includes a reduction in CMOS smear and a small improvement in dynamic range (.5 stops maybe) Gavin and I would like to actually measure EPIC Dragon, Weapon Dragon and Helium in a few weeks to show some of the differences. In my personal opinion, Weapon Dragon is what EPIC Dragon should have been or what I think we were hoping it would be. That is not to downplay EPIC Dragon at all. I just think they were able to improve it once enough were in the field and they could collect feedback.

All of the specs in the world do not matter when compared to real-world testing. If you were in WA State, I would let you test my Weapon Dragon VV.

 

[Antony Newman]

Stephen,

Another (in my case unexperienced!) perspective.

It would be 'nice' to have an absolute signal to noise ratio - but the reality is that the number would vary by a large number of parameters.

SNR probably depends on:
+) If R,G, Or B pixels (or a combination) were being used as input
+) The lumen light level
+) The freqencies of light on any RGB pixel (i.e. Some colours will be great, others less so)
+) The output resolution (oversampling)
+) Averaging (using higher frame rate for successive samples)
+) Current temperature
+) Mains (with Supply contamination) or Battery
+) Qualty of Lens
+) 'OLPF'
+) Raw signal processing (and artifact removal)
+) Redcode Compression
+) Post processing (e.g. ADD, DEB)

There are probably others!

AJ

 

[Phil Holland]

The short answer comes from comparing images side-by-side from both cameras.

You sincerely can't judge any camera just on printed specs alone.

In terms of RED advertising DSMC2 benefits, they've been doing that for a year and a half plus. Trade shows, online, print media. Jarred posted this even:
http://www.reduser.net/forum/showthread.php?139164-DSMC-2-The-Weapon-The-Scarlet-W-and-the-Raven-and-the-improvements-that-they-bring

I've personally done a host of tests and written down my thoughts.

S/N isn't the only thing that matters when it comes to the recorded image

 

[gabrielB]

I just a lowly REDUSER, but I owned an EPIC DRAGON and a SCARLET DRAGON before upgrading to two Weapon MGs and buying a Weapon W for good measure. The image quality of DSMC2 DRAGON vs. DSMC DRAGON is instantly noticeable. [...]

Michael thank you for your write up and for having taken the time to share that with us. Very informative!

 

[Blair S. Paulsen]

1) Measured stops vs usable stops:

The DSC Labs Xyla test chart Stacey notes is a common reference point for DR. A given sensor may have a native mathematical signal to noise ratio, but delivered system performance is the more useful metric. Moreover, the distribution of noise in the bottom end counts as well. In the early days of Epic Dragon, in underexposed footy, I noticed more noise in Zones 2 & 3 than I'd been seeing in MX. Zone 1 noise rarely bothers me much, as in most shots I can crush it out in the grade with minimal impact on the overall results. Noise that shows up in Zones 2 & 3 - which was reduced quite a bit on Dragon by the evolution of Graeme's color/gamma formulae, the superior processing on the DSMC2 bodies and (depending on which body) higher max data rates - typically requires a more intentional strategy to avoid causing issues. YMMV.

2) Tolerance for noise based on character/taste/project:

Some material can "carry" a moderate amount of noise and it just plays as texture. On other material, moderate noise is distracting and works against the desired look. There's even some material where moderate noise adds an organic feel that fits the creative intent better than clean - many a digitally acquired piece has had "grain" added in post by choice.

Telling stories with imagery has always been a synthesis of science and art, noise/grain is one of many elements where they are interdependent.

Cheers - #19