Noisy blacks in Epic as compared to new Dragon sensor.

[JR Monteiro]

I have been reading all of the posts regarding how to deal with noise in low light situations i.e.. the importance of doing black shading at the frame rate you will be shooting at etc.. But here is my problem with all those strategies, and what I have experiences shooting the Epic myself. When there is a dark area in the frame that I want to keep dark regardless of the lighting situation (i.e.. a black car, the opening into a cave or a doorway to a darkened room) my noise meter alerts me to the increased noise in that underexposed area. Yes I can crush the blacks in post as suggested by some but not if I want to keep detail (such as in the black car in the sunlight). It doesn't matter what ISO I set the camera at, all pixels on the sensor exposed within 10-15% of pure black, based on the native ISO of the sensor will have noise in them. This is very different from say the Cine Alta HD cams I used to use. When you switched on the gain, 9db or 18db the noise would appear evenly across the full frame whether it was under or over exposed. At 0db gain you could cap the lens and the blacks would be relatively clean. Capping the Epic lens gets a very different result. For example, I tried shooting a time lapse sequence of clouds passing across the night sky in front of the moon and the vast dark areas of the sky were noisy, even after doing a black shading. I have done similar tests with a 5D and never noticed the noise. What is different between these two systems? And does the Dragon sensor take a different approach? From the posts it seems that changing the ISO setting will actually affect more than just the picture on the monitor.

 

[Jason Comparetto]

I'd love to know what other people do about this same problem. Have had to crush the blacks before... ALOT. Not always happy with the results

 

[luigivaltulini]

I have been reading all of the posts regarding how to deal with noise in low light situations i.e.. the importance of doing black shading at the frame rate you will be shooting at etc.. But here is my problem with all those strategies, and what I have experiences shooting the Epic myself. When there is a dark area in the frame that I want to keep dark regardless of the lighting situation (i.e.. a black car, the opening into a cave or a doorway to a darkened room) my noise meter alerts me to the increased noise in that underexposed area. Yes I can crush the blacks in post as suggested by some but not if I want to keep detail (such as in the black car in the sunlight). It doesn't matter what ISO I set the camera at, all pixels on the sensor exposed within 10-15% of pure black, based on the native ISO of the sensor will have noise in them. This is very different from say the Cine Alta HD cams I used to use. When you switched on the gain, 9db or 18db the noise would appear evenly across the full frame whether it was under or over exposed. At 0db gain you could cap the lens and the blacks would be relatively clean. Capping the Epic lens gets a very different result. For example, I tried shooting a time lapse sequence of clouds passing across the night sky in front of the moon and the vast dark areas of the sky were noisy, even after doing a black shading. I have done similar tests with a 5D and never noticed the noise. What is different between these two systems? And does the Dragon sensor take a different approach? From the posts it seems that changing the ISO setting will actually affect more than just the picture on the monitor.

ciao JR,
many times people do the render in the middle 'debayer, you have always done to render the full debayer?
that compression shots often seen? 8:1? 3:1?
you can post some fragments grabs, here on reduser, I'm sure someone can help you.

Luigi

 

[Alan Peterson]

Flat black/dark areas where shadow noise shows up the most can take a lot of noise reduction without qualitatively degrading the image. It's probably worth playing around with a noise reduction filter that can single out a specific luma range, like the Boris one, to see if it does what you want. Other than that, I've had no problems to begin with when shooting at low ISO settings (if I know I'll want to fish a lot of detail out of shadows, I usually turn it all the way down to 250). High contrast lenses with modern coatings can help make your shadows less muddy, too.

 

[Zhibo Lai]

I'm relatively new Red shooter myself, 1+ years.. By no means an expert, but i've been happy with my blacks lately.

First, are you getting these "noisy" levels in your final delivery or just during preview in say Redcine-X? If you view your footage in Redcine-X at anything less than full debayer, it will appear more noisy than it actually is. At first I thought I just had a noisy camera since my Redcine defaults to 1/8 debayer, but upon switching to full debayer most of the apparent "noise" in the blacks goes away and I'm happy.
If you want super clean blacks, then Neat Video is your friend. This program has saved my butt many times over when I needed a clean image or just didn't have sufficient control of lights at the time.

The other thing to note is that Red cameras do not do any noise reduction, that is something left up to post do decide on applying. Cameras like the 5D apply noise reduction and compression for you. Both of these things will kill a lot of the noise. Difference here is that you don't have a choice what noise reduction or compression to use, with Red you do. In noise reduction physics, when you downscale or apply compression, you reduce noise. Red leaves those options up to you in post. The other cameras do not and force their crappy compression schemes on you, thus they "appear" to be less noisy.

 

[Mike P.]

I think your Cine Alta reference is very telling; if you're at "0db" gain or base ISO, most (all?) other camera systems reproduce black as clean black. Not with RED. It's frustrating as it requires work-arounds and using tricks to get the proper image when it should otherwise be simple. Dragon should be different, since the amount of dynamic range stays the same at any ISO. So at ISO100 you should detail in 3 stops lower in the blacks than ISO800 (which will have info 3 stops higher). It's a hot concept and what other cameras (Alexa, Canon C-series) already have, but MX does NOT have. Lesson learned, I guess.

Some tricks I've tried are using a low-con filter to essentially raise the blacks and then using lift/shadow to drop them back down to 0ire. It works okay (since it doesn't necessarily crush the black detail, so much as bring black back to black) but the rest of the image could suffer as a result (get muddy), so a lot of times it isn't worth it. On the same token, adding fill light sometimes isn't practical/possible either, so this can be lesser of two evils.

Another thing to consider is that the sensor's "native" ISO is closer to ISO200-320 than it is ISO800. 800 is recommended because it offers a "clean" image from a signal-to-noise standpoint, but I don't think it's the sensor's default. While some would argue that there is "no native ISO as it's meta data", that's actually kind of BS because regardless of the ISO the sensor clips and crushes at fixed points (if the goalposts are triggered, they remain the exact same regardless of what the ISO is... So the sensor does have a native ISO, it's just that mid-gray can be changed... It's sensitivity cannot be changed.) In other words, the RAW captured image is ALWAYS exactly the same... and whatever "ISO/ASA" brightness that capture is, is what the sensor actually is (and presumably where black would be clean black regardless of how much light is in the frame.) You won't get an honest answer since it's probably really low (ISO200) and because ISO800 is technically "clean."

OH, and REDcode compression adds noise. Compare it to an externally recorded image (Pix, Samurai, etc.) and the external image acts more like the traditional camera in terms of gain/ISO and noise. Conversely (and somewhat ironically), the blacks there are usually "clean" at ISO800 (even when you put the lens cap on).

Edit: This is all my own conjecture; no science to back it up, just what I've seen/tested in my own footage.

 

[Alan Peterson]

I think your Cine Alta reference is very telling; if you're at "0db" gain or base ISO, most (all?) other camera systems reproduce black as clean black. Not with RED. It's frustrating as it requires work-arounds and using tricks to get the proper image when it should otherwise be simple.

Most other camera systems have degraining filters built into their firmware. Shooting RAW doesn't allow for that, so it's not really an apples-to-apples comparison. Think of it like shooting RAW vs. JPEG on a DSLR - scroll through the high-ISO tests on dpreview or a comparable site, and you'll see that the JPEG images come out cleaner at the same ISO on the same camera. I'd like to see a cleaner ISO 800 on the Epic too but, just like in the still photo world, noise reduction is a normal part of RAW processing. I'd love to see luma noise reduction built into RCPX, though...

 

[Mike P.]

As far as I'm aware, they don't apply noise reduction to the HDMI/SDI out, so that whole "everything but RED has noise reduction" doesn't hold that much weight since the blacks are cleaner there (and not just because of the downscale). Also all other RAW cameras -- F65, Alexa, C500, even the 5Dmk3 with ML -- don't apply noise reduction, or if they do their images only benefit from it. In fact, there is no real-world case where applying noise reduction harms the image so much that you'd actually turn it off.

Case in point, having to apply 3rd party noise reduction to every frame of your RED footage in order to clean up the blacks. That's horrible logic and no amount of retro-justifying will make it a "better" process. It should be ON be default with the option to turn it off. And I'd be willing to bet that 99.999% of productions would never turn it off because the image would look noticeably better and it'd save time compared to being forced to do it in post anyway.

Don't get me wrong, I like RED cameras, but even they admit that they don't know what they're doing. This is one of those things where they had a brain-fart (the theoretical reasoning doesn't come close to the practical one)... same with the cooling fan design. Whoever thought fixing it in post was better than supplying an onboard option made a mistake (which they seem to have rectified on Dragon... theoretically anyway).

EDIT: Also, when put into context, it isn't nearly as bad as I'm making it sound. I mean, RED footage at ISO800 is cleaner than any film stock directly out of the gate. I'm just mentioning this stuff because, like the OP, I have noticed this phenomenon and it's just a side effect of the way RED cameras do things.

 

[Alan Peterson]

Where are you seeing this Alexa/F65/C500 RAW footage with no shadow noise? Productions shot on those cameras go through degraining in post just like RED cameras...

 

[Joesef o.]

I've had many issues with the blacks when I first got my RED until I realized the base ISO is NOT 800 but around 320. I now light and shoot for ISO 320, and get lovely clean blacks. I also take the shadows and knock them down a tiny bit which really helps too. Not crushing blacks with contrast or anything.

Also I shoot around 6:1 which is much nicer looking, in my opinion vs 8:1 (which a lot of people shoot at to save data)

6:1 gives me what I like. I'm a noise Nazi so believe me it is annoying. I came from a Sony Cine Alta background as well. The blacks did seem cleaner, but honestly that was with 1/4th the resolution. I'm pretty sure that if they were 4K cameras, we'd see the noise more as well.

Light everything for 320 is the best I can tell you, I shot a feature earlier this year and overlit my set so that 320 exposed pretty well. The images I got were so clean and organic, I loved it. Shooting without exposing the MX sensor properly just introduces all that noise. Don't listen to anyone saying its native 800... thats a wash, I can't even use 800 no matter what I do unless the scene already has a HUGE amount of light. (then its pointless to use 800 anyway).

 

[JR Monteiro]

Thanks guys. I am in agreement that stuff I have shot at 320 looks better than 800 ISO especially when I am shooting Tungsten. I made the mistake of doing Green Screen under tungsten at 800 ISO and almost had to do a reshoot there was so much blue noise in the green screen background. But again whatever the native ISO 320, 200, 800, the black areas of the shot will show identical noise. Granted it is not that bad as compared to some other cameras and it only shows up in the black areas but it is definitely disconcerting when you see that noise bar at 100%. Dealing with the noise in post after the fact is one thing I just want to make sure I am going into post with the best image I can. Exterior night shots are a good example when only small parts of the frame are fully exposed. No matter what ISO you chose the noise bar is going to read 100% unless you flood the shadows with so much unnecessary light that you get rid of the dark areas altogether and thus destroy the mood you are trying to create. I guess the answer is just to use proper exposure techniques and ignore the noise bar realizing that it is not telling you how noisy the picture is but how much of the picture area will have noise in it. Does this make sense?

 

[Joesef o.]

Thanks guys. I am in agreement that stuff I have shot at 320 looks better than 800 ISO especially when I am shooting Tungsten. I made the mistake of doing Green Screen under tungsten at 800 ISO and almost had to do a reshoot there was so much blue noise in the green screen background. But again whatever the native ISO 320, 200, 800, the black areas of the shot will show identical noise. Granted it is not that bad as compared to some other cameras and it only shows up in the black areas but it is definitely disconcerting when you see that noise bar at 100%. Dealing with the noise in post after the fact is one thing I just want to make sure I am going into post with the best image I can. Exterior night shots are a good example when only small parts of the frame are fully exposed. No matter what ISO you chose the noise bar is going to read 100% unless you flood the shadows with so much unnecessary light that you get rid of the dark areas altogether and thus destroy the mood you are trying to create. I guess the answer is just to use proper exposure techniques and ignore the noise bar realizing that it is not telling you how noisy the picture is but how much of the picture area will have noise in it. Does this make sense?

hmm now I find it interesting you still see blacks if you are rating for iso 320 and viewing in iso 320, you should have no noise. Should be nice a smooth blacks for the most part. I've only had issues seeing noise if I under-exposed 320 and had the iso at 640 thinking it was ok. Check and shoot looking at the RAW image on the monitor and you should be fine. In RCXP you should view again at ISO320 and there shouldnt be much, if any noise.

Take the shadows down a little bit too, this smooths that out a bit, but I'm pretty sure I didn't have any noise blacks so long as I exposed for 320 and VIEWED it at 320. Even very very dark scenes, so long as I view at 320 it wasn't super noisy. Can you post R3D snap shots to show what you're talking about?

 

[emil tanem]

I concur with the 320 iso experience.

I've done several comparisons where I've seen a cleaner image from the Epic than the Scarlet (in the blacks). I now have my own Epic and I'm still seeing this difference. This is on exact similar settigs. As far as I'm aware this doesn't make any sense technically. It could of course have been something slightly wrong with my Scarlet.

 

[JR Monteiro]

Hi Josef, The noise dramatically improved when I did a black shading as suggested on another thread, however; although the noise is not necessarily objectionable, what I do see is the little red noise level meter (just to the left of the histogram) go up and down as I pan into and out of dark areas or have a large portion of the frame comprised of black or dark shadows. Frankly it got me worried. I thought that I was doing something wrong. Now I am beginning to understand is that the meter is not a measure of how noisy the shot is but what percentage of the shot is comprised of black pixels where the noise resides. Correct me if I am wrong but this mean that any shot comprised of mostly black pixels is going to read 100% on the noise meter regardless of ISO setting. This is totally different then the ENG video cameras I used to work with. These cameras were rated with a measurement called "Signal to Noise Ratio" The amount of noise in the signal regardless of how much black area there was in the picture. That noise was distributed through both the light and dark areas of the picture even though it was more identifiable in the dark areas because there was not detail to distract the eye from seeing the noise. As you "Pushed" the sensitivity of the camera 9 or 18db, the noise became more prominent throughout the frame. With the Epic, as I understand it now, the noise resides only in the under exposed portion of the screen and no where else. So I am guessing that any noise reduction filters such as "Neat Video" that some of you have recommended, really only needs to be applied to the dark areas of the image since there is really no noise in the other areas. I would be interested to know if there has been a test done of the "Signal to Noise Ratio" of the Epic with a capped lens providing pure black. I like to shoot night-for-night with large areas of black in the frame for dramatic effect I guess I shouldn't fear doing this despite what that little red noise meter tells me.

 

[Alan Peterson]

The product info page lists the signal to noise ratio as 66db:
http://www.red.com/products/epic-mx#tech-specs

Also, I'm pretty sure the "red noise meter" you're seeing is just an indicator to show your blacks are clipping. The more of your frame is in absolute black, the higher that meter goes. There's one for whites on the right of the histogram, too.