Playing with the Forged Carbon Fiber Weapon and it's AWESOME!

[Zeb B]

Been traveling a bunch with limited access to interweb. No time to sort through RU to see if there are any R3D's posted from field users yet. Anything up yet?

Are the the couple R3D's posted on this thread the only ones anyone has uploaded?

Is it just me or does RED shadow noise have a very distinct DNA in that it creates colors that are not there? Why not just neutral luma noise . . . what's with the big grain chroma noise? I'm guessing her shirt was black and skin was pink.

So was just reading a review from the boyz at Lens Rentals:
http://www.lensrentals.com/blog/2015/08/sony-a7r-ii-a-brief-review

And this came to mind again. Why is it that the RED cameras color the noise instead of just straight luma noise like the Sony A7 series? The luma noise below is just grain to me - not a random color array. Here's an A7RII shot at ISO 12K with the 100% crop below. It's was a 42MP still. Not trying to do anything but understand the underlying challenge or color science approach or limitations that create this phenomenon and to know if Weapon is going to be real different in this regard than Dragon.

100% crop showing shadow detail:

Links to files:
http://imageshack.com/a/img913/3859/TAh2Wt.jpg
http://imageshack.com/a/img905/7798/MCkGSe.png

 

[Tarquin Cardona]

One thing that leaps out is how red/pink the guy's face is. Also if I remember correctly the crop in on the woman in the Epic picture is of a way smaller area in the frame, and also not a 42megaP still but a motion capture image of much smaller dimensions so the magnification element can come into play. But anyway. I have observed this phenom before and I think Graeme even responded. I actually felt it was part of some secret sauce on Red's part. Not a limit.

The thing to me about Red is that there is a special realism and a color density they get that is missing in almost every other camera. I have 'amateurishly but earnestly' attributed it to the fact the camera sees life in more complex color than a lot of other cameras, like a Sony F5 for example or C300. Skin is complicated and it took Red a long time to nail that. All objects are receiving, absorbing light and reflecting it back. Some direct from a source, some bounced from a nearby wall or someone else's shirt or whatever. So reality is much more complex than basic skin tone, color charts or black and white. Some camera companies opt to process their images to simplify things. So the image looses density, realness, but the colors behave more cleanly. Black is black. White is white, and the face is more a uniform 'skin' look. The result is that the picture is perfectly fine, totally acceptable, good enough. But to me often, especially with video, just not as real - and not as absorbingly rich. Just one persons opinion of course.

 

[Zeb B]

One thing that leaps out is how red/pink the guy's face is. Also if I remember correctly the crop in on the woman in the Epic picture is of a way smaller area in the frame, and also not a 42megaP still but a motion capture image of much smaller dimensions so the magnification element can come into play. But anyway. I have observed this phenom before and I think Graeme even responded. I actually felt it was part of some secret sauce on Red's part. Not a limit.

The thing to me about Red is that there is a special realism and a color density they get that is missing in almost every other camera. I have 'amateurishly but earnestly' attributed it to the fact the camera sees life in more complex color than a lot of other cameras, like a Sony F5 for example or C300. Skin is complicated and it took Red a long time to nail that. All objects are receiving, absorbing light and reflecting it back. Some direct from a source, some bounced from a nearby wall or someone else's shirt or whatever. So reality is much more complex than basic skin tone, color charts or black and white. Some camera companies opt to process their images to simplify things. So the image looses density, realness, but the colors behave more cleanly. Black is black. White is white, and the face is more a uniform 'skin' look. The result is that the picture is perfectly fine, totally acceptable, good enough. But to me often, especially with video, just not as real - and not as absorbingly rich. Just one persons opinion of course.

I get the mixed available light and grade is shifting the skin tones. It's the noise that I'm trying to understand. By digital definition, Black has no RGB information. Here's a 400% Crop:
http://imageshack.com/a/img661/3641/ZHETpZ.png

 

[Patrick Tresch]

I agree with you about the essencial (kill noise) but:

There is no such a thing as luma noise in an untouched raw file as the picture is based on R G and B pixels. So you can blur out the saturation noise like on a simple node composition in resolve but you are "chaeting" or post-processing the raw negative. I think RED thinks that it has to be managed in post by competent people with the appropriate tool and they don't want to make the trick a ready made burnt in effect. Perhaps they could implement this kind of trick for the prores encodes?

Pat

 

[Tarquin Cardona]

Patrick. I seem to recall that being Graeme's response to my 'theory'. Red does not tweak, process or simplify the image in that way. But we have become so sensitive and attuned to issues like CA that we can interpret these things as say color pollution even if it is maybe more accurate color capture.

It would be interesting to know if they could devise a snappy pro rez encode that pulled the same trick if one wanted it.

 

[Phil Holland]

So was just reading a review from the boyz at Lens Rentals:

Here's an A7RII shot at ISO 12K with the 100% crop below.

Sony for a while now has been employing a very specific type of noise reduction that virtually eliminates chromatic noise via an averaging and blurring technique. While the high ISO is impressive, there is very little chromatic or tonal separation in the image.

The calling sign for this method is usually the sandpaper texture which reveals itself as tonal/luminance noise.

RED cameras do not utilize in camera noise reduction techniques. At least not yet. One of the big reasons Canon motion picture cameras haven't penetrated "certain" fields is due to the fact they have internal noise reduction. Most Digital Cinema style cameras should not and do not have NR. Especially if you are dealing with something closer to true compressed RAW data. NR on RAW data is simply not RAW.

 

[Charles Bergquist]

Does the Alexa in any form apply noise reduction?

 

[Nick Morrison]

Does the Alexa in any form apply noise reduction?

Traditionally no.

But since their UHD functionality has been creating additional noise, I believe they've been offering in camera NR to compensate.

UHD appears to have philisophically challenged Arri - they are now upconverting AND adding NR!

Then again, Arri appears to have philosophically challenged RED, who are now providing Prores in camera, and no longer just shooting RAW!

 

[Zeb B]

I'm not real clear on why any form of in-camera Noise Reduction is a bad thing. This has been the mantra and accepted POV on this forum. Seems like an elegant algorithm with a slider to adjust the noise removal would be best coded by the same team that writes the RC code. They know the best approach to clean up the image better than Neat Video or whoever. It's not just higher ISO / low light noise - it's also the noise from higher RC compression than would be cool to address in-camera. Could save an entire step in the post processing workflow. From a time management perspective, applying Neat to a big rez timeline is an invitation to run some errands, do the laundry or even repaint the house ;-)

Can we have RAW high tonal range and clean, honest shadow color out of camera? Why is it one or the other?

 

[Craig Lees]

Whaaa? Weapon ISO 5000 = ISO 1280 on Epic Dragon??

That is not a half-stop increase in s/n d/b ratio!

I would have put a deposit down on a Weapon if this was true at time of NAB marketing. That's like VariCam performance. Even if it is NR, this is big. How many times are you crippled by noise when shooting slow-mo on Red (especially at night).

Damn.

Are RED confirming how clean the images are at ISO5000? Is this caused by the super-sampling of the 8K weapon sensor or is this from the standard 6K Dragon sensor?

Interesting.

Craig Lees
ED UK
3dstories.co.uk

 

[Nick Morrison]

I'm not real clear on why any form of in-camera Noise Reduction is a bad thing. This has been the mantra and accepted POV on this forum. Seems like an elegant algorithm with a slider to adjust the noise removal would be best coded by the same team that writes the RC code. They know the best approach to clean up the image better than Neat Video or whoever. It's not just higher ISO / low light noise - it's also the noise from higher RC compression than would be cool to address in-camera. Could save an entire step in the post processing workflow. From a time management perspective, applying Neat to a big rez timeline is an invitation to run some errands, do the laundry or even repaint the house ;-)

Can we have RAW high tonal range and clean, honest shadow color out of camera? Why is it one or the other?

Well, there's clearly a reason the three most accepted and commonly used cameras in CINEMA (ALEXA, RED EPICS, SONY F65) DON'T denoise in camera.

1) It's considered far better to handle denoising in POST. A colorist on a premium grading platform has far more control over the look and denoise algorithms than a "computer" doing it on the fly, as you shoot, in a manner that's destructive and irrevocable.

2) Skintones. Because denoising algorithms attack the RGB channels on the fly, often the RED channel can be adversely affected, which can skew skintones and accurate color reproduction.

3) Sharpness. Denoising can whither the RED and BLUE channels, and cause noticeable declines in sharpness.

In general, it's widely considered better to denoise in post. This is not a "RED" thing. This is also widely accepted in the Arri community, where there cameras, until recently with adaptation of UHD, have never denoised in camera either.

Notice the F65 doesn't denoise (the flagship cinema camera), whereas the F55/5 do.

 

[Zeb B]

Well, there's clearly a reason the three most accepted and commonly used cameras in CINEMA (ALEXA, RED EPICS, SONY F65) DON'T denoise in camera.

1) It's considered far better to handle denoising in POST. A colorist on a premium grading platform has far more control over the look and denoise algorithms than a "computer" doing it on the fly, as you shoot, in a manner that's destructive and irrevocable.

2) Skintones. Because denoising algorithms attack the RGB channels on the fly, often the RED channel can be adversely affected, which can skew skintones and accurate color reproduction.

3) Sharpness. Denoising can whither the RED and BLUE channels, and cause noticeable declines in sharpness.

In general, it's widely considered better to denoise in post. This is not a "RED" thing. This is also widely accepted in the Arri community, where there cameras, until recently with adaptation of UHD, have never denoised in camera either.

Notice the F65 doesn't denoise (the flagship cinema camera), whereas the F55/5 do.

Just sounds like the technology isn't there yet . . . like driver-less cars 3 years ago. All your points could be addressed with proper coding and user adjustable parameters. Basic NR would make Dragon come out of the bag more often since not every shot is 24p RC 2:1 lit with 20K's. Also, RED style noise is unique to RED and is always creeping into frame.

Dealing with noise and random color in the blacks is another workflow step that adds time and cost. Therefore that step will be eliminated since that's how Capitalism works.

Also, never did agree that "just crush the blacks to get rid of the noise" is an effective method to preserve tonal range. It's a fix, but not every shot is a Hollywood blockbuster preset look. Perhaps the subtle gray tones in the folds of a black fabric shot available light and low key is desired. For example, the above shot of the grooms pants.

Five years from now we will look back to 2015 and say "WTF and why did we have to work so hard to get a clean image from expensive movie cameras".

 

[Patrick Tresch]

"WTF and why did we have to work so hard to get a clean image from expensive movie cameras".

Now this is TRUTH! ;-)

 

[Nick Morrison]

Just sounds like the technology isn't there yet . . . like driver-less cars 3 years ago. All your points could be addressed with proper coding and user adjustable parameters. Basic NR would make Dragon come out of the bag more often since not every shot is 24p RC 2:1 lit with 20K's. Also, RED style noise is unique to RED and is always creeping into frame.

Dealing with noise and random color in the blacks is another workflow step that adds time and cost. Therefore that step will be eliminated since that's how Capitalism works.

Also, never did agree that "just crush the blacks to get rid of the noise" is an effective method to preserve tonal range. It's a fix, but not every shot is a Hollywood blockbuster preset look. Perhaps the subtle gray tones in the folds of a black fabric shot available light and low key is desired. For example, the above shot of the grooms pants.

Five years from now we will look back to 2015 and say "WTF and why did we have to work so hard to get a clean image from expensive movie cameras".

I still think deciding in post is better than having choice made IN-CAMERA, irrevocably.

Also, have you seen this recent Amira vs Dragon test?

https://vimeo.com/135498367

Dragon noise doesn't seem that different than Amira/Alexa (the gold standard). And with Weapon NOW looking noticeably cleaner than Epic Dragon, Weapon may today actually equal (or perhaps surpass?) Alexa in noise, while still retaining extra sharpness and resolution, not to mention Dragon's gorgeous color, which in his test Harrison thought looked even better out the box than Amira.

So I think RED's heading in right direction.

From early tests, 5000 iso is looking pretty clean, even WITHOUT NR...so that's a very good sign.

 

[Kemalettin Sert]

I still think deciding in post is better than having choice made IN-CAMERA, irrevocably.

Also, have you seen this recent Amira vs Dragon test?

https://vimeo.com/135498367

Dragon noise doesn't seem that different than Amira/Alexa (the gold standard). And with Weapon NOW looking noticeably cleaner than Epic Dragon, Weapon may today actually equal (or perhaps surpass?) Alexa in noise, while still retaining extra sharpness and resolution, not to mention Dragon's gorgeous color, which in his test Harrison thought looked even better out the box than Amira.

So I think RED's heading in right direction.

From early tests, 5000 iso is looking pretty clean, even WITHOUT NR...so that's a very good sign.

do not care about noise anymore when you look at that test clearly 5 yo ALEV III sensor kicking some ass.

 

[Zeb B]

I still think deciding in post is better than having choice made IN-CAMERA, irrevocably.

Also, have you seen this recent Amira vs Dragon test?

https://vimeo.com/135498367

Dragon noise doesn't seem that different than Amira/Alexa (the gold standard). And with Weapon NOW looking noticeably cleaner than Epic Dragon, Weapon may today actually equal (or perhaps surpass?) Alexa in noise, while still retaining extra sharpness and resolution, not to mention Dragon's gorgeous color, which in his test Harrison thought looked even better out the box than Amira.

So I think RED's heading in right direction.

From early tests, 5000 iso is looking pretty clean, even WITHOUT NR...so that's a very good sign.

Explain that test like I'm 5 years old: How do you get Dragon's 16 or whatever claimed stops of DR if, in the test, noise has compromised color fidelity at -5EV and the skin starts clipping at +4.5EV. To my simple childlike brain that equals around 9 maybe 10 stops of usable DR.

 

[Satsuki Murashige]

Explain that test like I'm 5 years old: How do you get Dragon's 16 or whatever claimed stops of DR if, in the test, noise has compromised color fidelity at -5EV and the skin starts clipping at +4.5EV. To my simple childlike brain that equals around 9 maybe 10 stops of usable DR.

I think you are confusing dynamic range with exposure latitude. Dynamic range is the total number of f-stops captured by the sensor from highlight to shadow without clipping (objective) or unacceptable noise (subjective). You need to know how stops over/under your highlights and shadows are in the scene relative to middle grey to really know what the range of the scene is. Is the highlight on the c-stand knuckle 4 stops over? If so, then when does it clip in the series of overexposed wedges? That would give you the upper limit of the the range at a given ISO.

Exposure latitude is the acceptable range within which you can over or underexpose and still "print" back to normal. So in this example, let's say the usable range is -4.5 to +4, or 8.5 stops of latitude (based on your subjective taste of acceptable mid-tones). That's a lot, but it doesn't take into account highlights or shadows that may clip or get unacceptably noisy since you are just looking at mid-tones. If highlight and shadow detail is equally important to you (also subjective), then your latitude range may end up being quite a bit smaller.

 

[Filip Orlandic]

Explain that test like I'm 5 years old: How do you get Dragon's 16 or whatever claimed stops of DR if, in the test, noise has compromised color fidelity at -5EV and the skin starts clipping at +4.5EV. To my simple childlike brain that equals around 9 maybe 10 stops of usable DR.

EBU claims: "Therefore 15 stops should be regarded as the extreme limit, 14 stops or less delivering useful pictures. The dynamic range is about 15 stops, although the lowest part of the range is only marginally useful."

 

[Nick Morrison]

EBU claims: "Therefore 15 stops should be regarded as the extreme limit, 14 stops or less delivering useful pictures. The dynamic range is about 15 stops, although the lowest part of the range is only marginally useful."

Weapon should change that. As Dragon sensor looks considerably cleaner in shadows now...inside Weapon body...

 

[Filip Orlandic]

Weapon should change that. As Dragon sensor looks considerably cleaner in shadows now...inside Weapon body...

Half a stop cleaner shadows with Weapon...