Dragon Noise Confirmation

[Matt Ryan]

Looks good Evin. I played around with your R3D and was surprised how little noise there was even though it was shot under 2900k lighting and could have been opened up a stop more. I processed this shot at 2000ISO. I probably could have gotten even better results if I could have used A.D.D.

 

[Gavin Greenwalt]

That's because by always exposing to the right, you will have to constantly change the development side of the exposure/development relationship. If you you expose a dark scene to the right, you change where middle gray is, you will have to pull the exposure down in post, and that will put it onto a different part of the gamma curve.

Unless I'm misunderstanding you, this is definitely inaccurate. If you take the linear output from an R3D and process it down (ISO) and then apply REDGamma/REDLogFilm it would be exactly the same (but less noisy) than exposing it darker. I have the REDGamma curve as a Nuke node for my comps and I do this frequently. I also often shoot "safety plates" for clean plates when shooting VFX and the over/under exposed images match exactly (except for noise).

 

[Phil Holland]

Unfortunately I'm noticing a purple color shift at the edge of frame even on the 35mm ASPH Summicron. The 21mm 2.8 is even worse and probably only useable in 3K, unless you have a special post fix that can work wonders then this mount is probably best reserved for 50mm+ or high-speed 3K/2K work.

Yep. Noticed that when I did the coverage craziness for the M glass on Dragon. The 24mm f/1.4 looked surprisingly good though still exhibiting what you saw. Correctable in post for sure, do it often. I'm primarily interested in producing a compact fast glass kit if it all works out. My Cine-Xenar IIIs are my workhorses, but fast glass is still fast glass and I have a desire to flesh out a something a bit more interesting than my EF range at the moment.

 

[Matt Ryan]

Like David, Peter, Bjorn, Evin and the others have said, the image is quite a bit underexposed. I processed the shot at 4000ISO to get an appropriate look and levels. You will need to use some noise reduction but it will look fine. If you despise noise I'd always recommend shooting at a lower ISO and bringing in more lights or opening up your lens. Nothing worse than underexposing at 800ISO and then having to bump exposure in post. That's asking for trouble. I'd also say the slight out of focus image also isn't helping the image and the perception of sharpness vs. noise. I also think if you were going for a night, moody look, processing was still around 1600ISO. As you know windowing the sensor down will reveal the appearance of more noise, so shooting at 6K in this situation may further benefit the shot.

4000 ISO:

1600 ISO:

 

[Medavoym]

When you take an incident light meter reading at your subject - and it reads (for example) T5.6 - that means that assuming your light meter and camera are both rated at the same ISO - is you set the aperture to T5.6 you are making you subject as bright as 18% grey. Caucasian skin tone is approx. 1 full stop over 18% grey. So ... you want to open up one full stop to T4.0. There can be a wide range of skin tone brightness even within caucasian - but 1 full stop is an average that is widely used.

This is not a Dragon thing - it's a measuring light thing.

With all due respect Mark, I think you are making the same confusion again. The above is flat wrong.

What you say holds value when taking a REFLECTIVE light meter reading. Not incident. When taking a reflective (spot meter) reading of a face and it reads T5.6, yes, you open up a stop (usually).

However, when taking an INCIDENT light meter reading, you do absolutely nothing. The incident value is the equivalent of spot metering a grey card. So, if the incident reading tells you T5.6, then you (usually) simply set your aperture at T5.6. The face will be exposed properly.

Just to set the record straight if newbies are reading this. We don't want any confusion.

Thanks my friend!

 

[Medavoym]

Here is an R3d frame properly exposed at EI800, using a meter (Sekonic L308) and the histo. Look into the brown sweater and the dark areas of the camera, there is texture but it's really well controlled. As a personal choice I prefer skin tones at metered exposure most of the time, I know Mark likes to open up a stop and that's valid for a more "lit" look.
http://www.evingrant.com/pics/ADD

And here is a 100% split frame between a regular snapshot (top) and an ADD snapshot (bottom)//???? ....

Tiff version of the split... http://www.evingrant.com/pics/ADD.tif

When I project this image to 84"@10' I can't see any noise at all, in fact even the severely underexposed frame from the OP looks great. I think too many are blowing things up to 100% in RCX and looking into the low quarter tones and seeing noise, well DUH! If you blow up any other camera to the same magnification and look into the low quarter tones (especially if it's underexposed) you'll be disappointed. Luckily nobody views our work this way, even on a 4K panel or projector with a 4K signal you need to factor in proper viewing distance and display output.

I'm probably going to be raked over for this but I don't want a digital cinema camera with no noise/grain! Whatever Red has done with the Dragon the noise character is the most "Filmlike" of anything I've seen, period. In fact I would say shooting Dragon at EI2000 is probably about equivalent to Kodak 5219@ EI 500.

Here is a 100% crop of a 1080P DPX frame from CML of scanned Kodak 5219 (No NR) next to my EI800 Dragon shot at 6K 1:1 100% (not ADD)...

http://www.cinematography.net/Skin/skin-2.dpx

For me the Dragon is a replacement for FILM, and the texture of noise is one of the most important parts of that, you can have a noiseless image is you like by shooting at lower EIs or ETTR or ADD or NR but I for one prefer the cinematic look of EI 800-2000.

I am in COMPLETE AGREEMENT with you my friend.

Noiseless/grainless images are lifeless, artificial, fake.

There is already too LITTLE noise/grain with good digital cameras now. Remove it obsessively and you destroy the texture and the magic.

It's your choice. You can pixel-peep and noise-peep all you want, and spend dozens of hours removing the noise (ADD, Neat Video), or you can accept Dragon as it is (an amazing camera with no real need to de-noise) and shoot away happily.

This is for narrative shooting. Stills - that's a different matter.

 

[Gavin Greenwalt]

Noiseless/grainless images are lifeless, artificial, fake.

Then add noise in post. Noise is incredibly easy to add to digital images. Grain... a little tricky. But noise is a purely additive process.

 

[Dr. Sassi]

Noise in Dragon, I think this clears it for me: http://www.reduser.net/forum/showthread.php?113541-RED-DRAGON-Scores-Highest-DXOMark-Score-ever

Grain in Film, I loved it, but today I can't stand it if I watch a BluRay and the image has heavy grain. If I'm at a film-festival I enjoy it, if the content fits to it, but I don't need it anymore. It has become an aesthetically indicator for me that points to the 20th century ;o). Things have changed for me.

 

[Mike P.]

Have RED said if ADD will make its way to the SDK/third-party apps for motion rendering or will it end up like Alchemy group? Also, have they mentioned if it's limited to Dragon or could the same effect be used on MX footage (and if not, why not)?

 

[Stacey Spears]

Also, have they mentioned if it's limited to Dragon or could the same effect be used on MX footage (and if not, why not)?

Here is what Jarred said:

Only Dragon footage ( hence the name ).

A.D.D. needs the extra information that only the Dragon sensor provides.. including the invisible bits.

 

[Mike P.]

Fiddlesticks. No mention of ADD in the SDK, though? So there is still hope for that!

 

[Tommaso Alvisi]

Neat video, is cheap and works great.

[/IMG]

Paulo,
do you add a little bit of video noise together with NeatVideo to avoid the posterization effect?

 

[Paulo Emílio]

Yes, Tommaso, I normally grade and comp in After Effects and grain is always in the final juice and 16 bits per channel.

 

[Robert Ruffo New]

With all due respect Mark, I think you are making the same confusion again. The above is flat wrong.

What you say holds value when taking a REFLECTIVE light meter reading. Not incident. When taking a reflective (spot meter) reading of a face and it reads T5.6, yes, you open up a stop (usually).

However, when taking an INCIDENT light meter reading, you do absolutely nothing. The incident value is the equivalent of spot metering a grey card. So, if the incident reading tells you T5.6, then you (usually) simply set your aperture at T5.6. The face will be exposed properly.

Just to set the record straight if newbies are reading this. We don't want any confusion.

Thanks my friend!

You are correct. The problem with reflective readings (which Marc is accurately describing, his name-switch is probably a typo, because he was tired when he wrote that, to his due credit) is that they are skewed by object color - so a black object will meter darker than a white one, even though both would be correctly lit at the same level of light. Incident readings do not care what color your scene objects are - they just observe how much light is hitting an area of the scene.

 

[Robert Ruffo New]

I am in COMPLETE AGREEMENT with you my friend.

Noiseless/grainless images are lifeless, artificial, fake.

There is already too LITTLE noise/grain with good digital cameras now. Remove it obsessively and you destroy the texture and the magic.

It's your choice. You can pixel-peep and noise-peep all you want, and spend dozens of hours removing the noise (ADD, Neat Video), or you can accept Dragon as it is (an amazing camera with no real need to de-noise) and shoot away happily.

This is for narrative shooting. Stills - that's a different matter.

The real problem is that noise does not recompress well for digital broadcast, Web platforms, etc. Especially a problem with dark material.

 

[Nick Morrison]

The real problem is that noise does not recompress well for digital broadcast, Web platforms, etc. Especially a problem with dark material.

So well said. I work in broadcast full time, and I see this problem all the time. We have lots of shows shot on Alexa, many are clearly "pushed", ie have a bit of noise already when we get the HDCAMS, and by the time we encode them 8bit to cut the promos...and eventually spit out "social media" qts...they look absolutely terrible. The blacks are so noisy they look devoured by gnats.

I fight noise in our social media deliverables all the time.

 

[Mark L. Pederson]

You are correct. The problem with reflective readings (which Marc is accurately describing, his name-switch is probably a typo, because he was tired when he wrote that, to his due credit) is that they are skewed by object color - so a black object will meter darker than a white one, even though both would be correctly lit at the same level of light. Incident readings do not care what color your scene objects are - they just observe how much light is hitting an area of the scene.

Yeah ... sorry guys. Didn't mean to add any more confusion

But of course, what is actually funny is that I must confess that I'm currently digging rating the Dragon at ISO at 640 - taking an incident reading at the subject (caucasian face) - and then open up a full stop - which means I'm technically over-exposing a stop But looks awesome IMO. So that added to me confusing the post - but that's an example of a personal "exposure method" based on ones subjective testing of a stock (or sensor).

I'm sadly still waiting for an initial OLPF upgrade - when I get that I'm gonna start over from scratch and re-test everything and try to post some "exposure exercise examples" - and document exact light meter readings and methodology for choosing exposure, ect. - and hopefully by then we will also have a new firmware with the Exposure Meter functioning correctly outside of RAW view.

 

[D Fuller]

Unless I'm misunderstanding you, this is definitely inaccurate. If you take the linear output from an R3D and process it down (ISO) and then apply REDGamma/REDLogFilm it would be exactly the same (but less noisy) than exposing it darker. I have the REDGamma curve as a Nuke node for my comps and I do this frequently. I also often shoot "safety plates" for clean plates when shooting VFX and the over/under exposed images match exactly (except for noise).

You are saying exactly what I'm saying. My point is that for the play of a scene, consistent noise levels matter. For two reasons - the look of the noise itself, and the things that are done in development in response to the noise (or lack thereof).

 

[Zeb B]

Kinda related:
http://www.reduser.net/forum/showthread.php?113604-Black-Shading-Confirmation

 

[Robert Ruffo New]

Yeah ... sorry guys. Didn't mean to add any more confusion

But of course, what is actually funny is that I must confess that I'm currently digging rating the Dragon at ISO at 640 - taking an incident reading at the subject (caucasian face) - and then open up a full stop - which means I'm technically over-exposing a stop But looks awesome IMO. So that added to me confusing the post - but that's an example of a personal "exposure method" based on ones subjective testing of a stock (or sensor).

I'm sadly still waiting for an initial OLPF upgrade - when I get that I'm gonna start over from scratch and re-test everything and try to post some "exposure exercise examples" - and document exact light meter readings and methodology for choosing exposure, ect. - and hopefully by then we will also have a new firmware with the Exposure Meter functioning correctly outside of RAW view.

I'm sure I am not alone in very much looking forward to seeing your findings.