Dragon ISO 100-6400 + ISO 50-12800

[Phil Holland]

A good chunk of my day these days is spent answering Dragon related questions. Every time something interesting comes up I feel compelled to share it here, because other likely have the same questions.


Today I was asked about the Carbon Dragon from somebody interested in purchasing one next month, and specifically this bit was brought up from the Red Store Epic Dragon Carbon Page:

RED DRAGON sensors have a native dynamic range of over 16.5+ stops, giving you a huge advantage in challenging lighting conditions. An improved signal-to-noise ratio gives you confidence that your picture quality will be print and cinema ready from ISO 200 to 2000. Less filters, less IR issues, and less noise gives you more opportunities for perfect exposure.

So what the heck does that actually mean and say. Well, after some projection tests I have thoughts that I want to share. Print and Cinema ready from ISO 200-2000 means exactly that. A workable image at up to ISO 2000 without any need for noise reduction. This I can confirm is true.

Interestingly enough because each scene is different you may be able to get a bit more out of a shot. I have pulled things down to ISO 50 with great results and up to ISO 4000 with decent results. If there's light hitting something in frame and it's within your raw exposure dynamic range, it's likely workable. However, much like Mysterium-X, if there's simply no light hitting something and you attempt to lift that up to the extreme you will see noise.


I've made two example clips. Keep in mind there is no curve involved here, just straight REDcolor3 and REDgamma3. Everything is just "flat". I am curious where DRAGONcolor will get us.

Clip 1 exhibits a scene that's well within the total dynamic range of Dragon's REDCODE RAW. Because of this it can be pushed a hell of a lot. ISO 100-6400 show here. 213MB

http://www.artbyphil.com/temp/reduser/phfx_Dragon_ISOTest1.mov

Clip 2 exhibits a black shirt as well as an unlit corner in the upper right hand of the frame, which creates essentially a "shadow box". Interestingly enough, you do see noise at the higher ISOs, but there is clear object separation over there. Pay special attention to the left corner where a bit of light is actually hitting as even at ISO 6400, it looks pretty damn nice. 347MB

http://www.artbyphil.com/temp/reduser/phfx_Dragon_ISOTest2.mov

In a realistic grading scenario you would likely apply a curve or somehow pull those shadow areas back down. However, I have certainly lifted visible detail midtones and highlights in the ISO 4000/5000 range with this little short shoot in the grade. I do think the vast majority of shooters are going to play in the ISO 200-2000 range though and expose accordingly. This is actually revealing itself to be a much broader range than Mysterium-X. Holding those highs and pulling down is damn impressive. Mark Toia's first footage out showcased that pretty well.

That wide dynamic range and workable latitude comes in real handy in post. In this image I exposed the subject who is in shadow at ISO 800. In the grade you can see how I utilized the extra space on the ceiling and floor.


- click for bigger image


Well that's my weird "Dragon on my mind" midnight post. However, it is showing something new that folks would want to see. Even exposing at ISO 6400 with that level of noise you could use a slight bit of noise reduction for shadow heavy shots and make it screen ready pretty easily and yield the oversampling benefits from going from 6K to 4K in retaining detail.

 

[Kwan Khan]

Thanks for sharing

 

[paulherrin]

nice. one question i still have lingering in my mind is if i'd somehow get less noise if i did multiple passes (shadows/highlights) - as in, once the debayer is done to rgb, the grade potentially introduces noise/artifacts when lifting that wouldn't be there if processed at a different exposure. as in pseudo-hdr... when i first started thinking and working with red footage like 5 years ago that seemed like a good idea, but i think the consensus was that you could get it all in there, but now i'm second guessing that consensus because you're doing histogram stretching after the rgb conversion that otherwise you could just blend. now probably it's just an academic thing and there'd be no perceptual difference, but there's still stuff like banding to think about if you're really pushing values. i guess i should really test it out - that's just always been one of those nagging things i've never acted on. any thoughts?

 

[Tom Gleeson]

Something I still do not understand about the Dragon sensor is that if I select either 200asa or 2000asa do I see the same spread of dynamic range in the highlights and shadows at either ASA setting or is it behaving more like the MX sensor?

 

[Phil Holland]

nice. one question i still have lingering in my mind is if i'd somehow get less noise if i did multiple passes (shadows/highlights)

There's faster ways to get better noise reduction at the highest ISO values.

Something I still do not understand about the Dragon sensor is that if I select either 200asa or 2000asa do I see the same spread of dynamic range in the highlights and shadows at either ASA setting or is it behaving more like the MX sensor?

It still works the same way as Mysterium-X, although with a significantly larger dynamic range.

This is good reading on how ISO works with Red cameras:
http://www.red.com/learn/red-101/iso-speed-revisited

Since the total dynamic range remains unchanged, this effectively shifts the number of stops that fall above and below middle gray, and compresses or expands tones accordingly. At low ISO speeds, most of the dynamic range is below middle gray, whereas at high ISO speeds most of this dynamic range lies above middle gray.

That expansion of tones is where noise can come from and that's when you're on the higher ISO side of things within the shadows and darks.

The main advantages to using this ISO method rather than how other folks typically do it is to capture the maximum dynamic range a sensor has to offer without "destroying it" through pre-processing and also to avoid any clipping or crushing when adjusting ISO values later in post.

However, when compared to Mysterium-X the simplest way I can explain the difference is this:

Compared to the sensitivity and dynamic range of Mysterium-X (ISO 320-800 and 13-ish stops of dynamic range), Dragon produces 16+ stops of dynamic range at about 1.3 stops higher base sensitivity (ISO 800-2000). However, because of that new dynamic range you capture more highlight and more shadow information.

Right now, and without doing hard tests, I think it's the cleanest digital cinema camera on the market. Oddly enough one of the newer cameras on the market's native ISO is considerably darker than expected. Don't exactly know what's up there.

 

[Tom Gleeson]

Thanks Phil for clearing that up

 

[Marcos Montenegro]

As always Phil, very informative. I still have printouts of your frame guides; your contributions don't go unnoticed here.

Curious if you ever pushed the compression ratios shooting 6K to see what RED was talking about developing a more efficient compression algorithm? If so, when could you start telling the difference in IQ? How big are the less compressed files?

 

[Eric Santiago]

I cant wait for the Phil RED COMPLETE Man pages

I want hard cover version

 

[Brian Pascale]

I Can't afford it, and don't really need it, but boy, oh boy, I sure want to upgrade the old Scarlet to a Epic Dragon. Okay, first I'm getting a new Mac Pro, then I upgrade the camera.

 

[Phil Holland]

As always Phil, very informative. I still have printouts of your frame guides; your contributions don't go unnoticed here.

Curious if you ever pushed the compression ratios shooting 6K to see what RED was talking about developing a more efficient compression algorithm? If so, when could you start telling the difference in IQ? How big are the less compressed files?

Yep. In the .R3Ds you can see various compression ratios at play as well as multiple frame rates from 24-96fps.

The major thing I noticed were how much details seems to be held at higher compression ratios. There will need to be hard tests obviously shooting some high frequency detail to really observe what's going on, but it's noticeable right off the bat. I think the lower noise floor in the Dragon material effects this and it's pretty impressive.


Back to Tom's question. I get asked that ISO/FLUT question a lot and it's been awhile since I've made a useful graphic. This pretty much shows exactly what's going on in reference to Dragon's Dynamic Range and adjusting the ISO or FLUT when rating your midtones.



There isn't a real visible penalty on the lower ISO (-2 stop) adjustment shown here, although I imagine slight color distortion could potentially occur on out of range highlights on the top end. In my brief shooting experiences with Dragon this is where the sensor reveals some of it's more "pleasing character" as the highlight roll-off looks very organic. It has more of a "burning out" effect.

This graphic does show when at the highest ISO values why you could potentially see noise in the darkest shadows if there's no light hitting that area.


Which is essentially why in that second clip above there's visible noise at ISO 6400 in that shadow box region frame right, but no visible noise in the other corner frame left. Essentially because light is hitting it.

So to reference Gunleik here where "Dragon is more", basically if the sensor can clearly see detailed tones within the Raw View it's likely workable material. Which is why some have been saying it "sees into the darkness".

 

[Jake Wilganowski]

Phil, with MX, we can set iso to 320, switch to raw view, and the histogram looks very similar, so most people say that's the "native" iso of the sensor. At what iso does dragon exhibit the same behavior?

 

[Phil Holland]

Phil, with MX, we can set iso to 320, switch to raw view, and the histogram looks very similar, so most people say that's the "native" iso of the sensor. At what iso does dragon exhibit the same behavior?

Heh. Contested yet not contested point right there. It's what I would describe as is "around ISO 320". If you really look at it under a variety of lighting conditions you'll notice that it isn't exact. There's a bit more in there.

As others have said though, you'll be looking at about ISO 800 with Dragon if you're looking to emulate that Raw View. Again though. There's a bit more in there than that.

 

[Jake Wilganowski]

Thanks. Very interesting. Gonna be fun!

 

[Mike P.]

Graeme has also said that the 320vsRaw view is a left-over from the RED1... Realistically you'd have to look at them in linear (at whatever ISO) since that's what the camera actually records.

Phil, isn't this test less useful since there was plenty of ambient light in the recorded image (and ISO, as they say, is just meta-data)... That is to say, did you adjust the physical exposure settings in any of these clips, or are you merely riding ISO in RCXp (or am I missing something)? If that's the case (riding ISO, and not changing exposure settings) what exactly are you showing up? That the colour/gamma remain more consistent across the ISO range than the do with MX?

 

[Brian Boyer]

I'm a little confused by the notation on your videos.

You have "RED DRAGON - REDcolor3/REDgamma3 - With S Curve". Across from that you have "No Curve, No Nothing, Flat".

So, I'm not sure what "With S Curve" means.

 

[Phil Holland]

Phil, isn't this test less useful since there was plenty of ambient light in the recorded image (and ISO, as they say, is just meta-data)... That is to say, did you adjust the physical exposure settings in any of these clips, or are you merely riding ISO in RCXp (or am I missing something)? If that's the case (riding ISO, and not changing exposure settings) what exactly are you showing up? That the colour/gamma remain more consistent across the ISO range than the do with MX?

Well yes and no. This was exposed properly for the shot and I'm merely adjusting ISO here, but there's a variety of content across the frame. If you look at Clip 2 in the upper right hand corner that is indeed an area of "darkness". That should tell you a lot.

I'm a little confused by the notation on your videos.

You have "RED DRAGON - REDcolor3/REDgamma3 - With S Curve". Across from that you have "No Curve, No Nothing, Flat".

So, I'm not sure what "With S Curve" means.

Doh. Apologies. I left my burn in text from an S Curve test. I'll upload new ones without that typo right now. It'll take a second for them to show up.

 

[Phil Holland]

Typos fixed and new versions uploaded.

I've also added one more. This clip is ISO 50-12800.

http://www.artbyphil.com/temp/reduser/phfx_Dragon_ISOTest3.mov

Again, I rated my exposure for ISO 800. What I'm essentially demonstrating is the range you have to work with within a balanced exposure. There's a lot of possibilities here. A lot.


** edit. One last additional clip. This one has a rather distinct and large dynamic range as there's deep shadow content as well as well lit. ISO 50-12800

http://www.artbyphil.com/temp/reduser/phfx_Dragon_ISOTest4.mov

 

[Nick Wernham]

Thanks for the neat info and nice images, Phil!

 

[Kemalettin Sert]

6400 like 1600 on MX..almost no noise incredible

 

[Phil Holland]

6400 like 1600 on MX..almost no noise incredible

I should say that these are straight h.264 exports from Redcine-X Pro. I have made DPX and encoded ProRes versions of these and I have to say, and we all know this, that there's a noticeable difference in quality.

At 4K under some lighting situations you might be able to get away with ISO 6400. Clip 3 in a less lossy codec looks actually tremendous at IS 12800, there's noise in the shadows, but it's really not crazy noise. Which is insanity.