First Clip from Crossing the Line

[Nick Shaw]

How about Bad Taste? To me that's DIY film-making of the finest kind. If you watch the making of on the DVD and see how much effort PJ went to, pretty much entirely on his own, it's quite astonishing. It made me realize how lame my amateur film projects at the time were, and inspired me to get into the industry.

 

[Zane Roach]

Good taste made Bad taste. I've watched that making of many times. It's great because it was made not long after Bad Taste, so its not about Peter Jackson, the guy who made the Lord of the Rings and Kong... It's about Peter Jackson, the guy who made Bad Taste!

He took a gamble, spent four years shooting, a lot of his own money on a movie with wooden machine guns and wooden acting... Who'd have guessed it would pay off?

Zane
(quits day job and starts carving wooden machine gun)

 

[Weston Ford]

I don't think this artifact is one that the film guys can pick on. The mechanical shutter in a film camera tears (shears) the image, visible in panning shots is you freeze frame and probably affecting the look of the motion slightly.

So the CMOS sensor in Red has an artifact that makes it look like film? Damn!

I know it can make some visual fx work tricky, but seriously the Red examples being posted lately are under more scrutiny than people put on 35mm film! :blink:

i can't imagine how anyone would find this shear thing acceptable. I've never even seen it before in a film...so I don't see how it can be called film-like.

And i'm not a celluoid lover at all. I really think digital is the way....thats why this worries me.

 

[Zane Roach]

i can't imagine how anyone would find this shear thing acceptable. I've never even seen it before in a film...so I don't see how it can be called film-like.

And i'm not a celluoid lover at all. I really think digital is the way....thats why this worries me.

Okay, I had another look at the clip and can spot it on straight vertical objects during pans. I have seen tearing in films, but only when freeze framing and looking for it. It may be more pronounced here, so I'll quiet down on the topic.

 

[Noah Yuan-Vogel]

This rolling shutter discussion is interesting, as is the posted clip, because i cant recall seeing this artifact when I saw the movie projected at NAB. I spent much of the screening leaning and squinting at the screen looking for imperfections, especially cmos-related ones (banding noise, rolling shutter artifacts, dynamic range limitations, etc.) Perhaps it was partly difficult to notice because of the fake camera shake (which if it were real might result in vertical compression or horizontal wobble because of the rolling shutter). I definitely noticed the banding noise in one of the shots, but it was later mentioned that that shot may have been pushed 3-4 stops in post.
I'm not sure what to think. Did anyone else notice this effect in the screening?

Unfortunately, what worries me is that RED hasnt come out and said "yes we have a great sensor that captures very true images without any distortion or artifacts" they only say something like "just trust us, it will *look* great, it wont necessarily look like film or like cmos with rolling shutter" i appreciate the need for secrecy, but its tough to make purchasing decisions and it's tough for those who already put down money when simple things like this arent addressed directly, especially for those of us in VFX...

 

[Joe Vinson]

I remember the Bad Taste making-of doc. Didn't P.J. make his own steadicam from scratch? Amazing guy!

 

[garageman]

It was more a "stabiliser" not a steadicam, big difference.

 

[Robert Jackson]

It was more a "stabiliser" not a steadicam, big difference.

Heh...wasn't it some kind of screen door hinge or something? I seem to remember it actually looking more like something he associated with a Steadicam than actually being a functional device.

 

[Steve Connor]

This rolling shutter discussion is interesting, as is the posted clip, because i cant recall seeing this artifact when I saw the movie projected at NAB. I spent much of the screening leaning and squinting at the screen looking for imperfections, especially cmos-related ones (banding noise, rolling shutter artifacts, dynamic range limitations, etc.) Perhaps it was partly difficult to notice because of the fake camera shake (which if it were real might result in vertical compression or horizontal wobble because of the rolling shutter).

From what I can gather no-one spotted it, so If no-one saw it on a projected 4K Film then how come it's been spotted on a compressed for the web clip? Does common sense not say it's in the web encode that the problem lies.

 

[garageman]

From what I can gather no-one spotted it, so If no-one saw it on a projected 4K Film then how come it's been spotted on a compressed for the web clip? Does common sense not say it's in the web encode that the problem lies.

..because people weren't going through the 4K clip frame by frame like they could at home with the 1K. A codec wouldn't cause this characteristic, it doesn't bother me too much anyway.

 

[Simon Blackledge]

As long as you can track, auto or by eye.. something centre frame on a pan, and you know the rate of distortion it should be possible to get a warper to match composited elements. Slower less, faster more. Not an easy task but very doable compared to some tricks plugin makers are doing these days.

Actually you could rebuild the pan to take the distortion out from previous frames.

Lots of "yeah but's.." like you have to start a pan so you get the plates/frames .. need to film more to one side to begin with etc... but just a thought. There's always options...

s

 

[Zk2007]

From what I can gather no-one spotted it, so If no-one saw it on a projected 4K Film then how come it's been spotted on a compressed for the web clip? Does common sense not say it's in the web encode that the problem lies.

So why RED doesn't come out and put this all to rest and say there's no problem with the real footage and it' being cause by the compression? I would love to hear that.

As long as you can track, auto or by eye.. something centre frame on a pan, and you know the rate of distortion it should be possible to get a warper to match composited elements. Slower less, faster more. Not an easy task but very doable compared to some tricks plugin makers are doing these days.

Actually you could rebuild the pan to take the distortion out from previous frames.

Lots of "yeah but's.." like you have to start a pan so you get the plates/frames .. need to film more to one side to begin with etc... but just a thought. There's always options...

s

Yeah but we are talking about painful workarounds and I was hoping workarounds wouldn’t be in RED’s dictionary. I was hoping it would just work. We have been going through workarounds for years with the low end stuff.
I just wonder the reason they designed the sensor that way. A global shutter is undoable at this size and or pixel count or what’s the reason?

 

[Robert Jackson]

So why RED doesn't come out and put this all to rest and say there's no problem with the real footage and it' being cause by the compression? I would love to hear that.

Would you? Cause that isn't an artifact that can be caused by any kind of compression I've ever seen. It's pretty easy to fix something like that, though. Calm down. It's not the end of the world. I'm sure the RED team are checking out the sensor thoroughly so they can make an authoritative statement on the issue when they break the silence.

 

[jbeale]

about the motion shear; remember that any motion picture film camera that has a physically moving or rotating shutter (I think that's all of them) has essentially the same behavior. So it's not a "new" problem. It's pretty much just the same behavior as we've always had, although it is actually easier to fix the Red-type shear in post, than the more complex behavior of a rotating shutter as used in 16mm or 35mm cameras.

 

[Álex Montoya]

That's a frame grab of the 1920*816 trailer of Bourne Ultimatum.


Jus in case anyone is losing perspective here.

 

[Ace]

Concrete,

That says it all.

And thats probably downsampled too..

 

[Frank Jonen]

shutter or not, could these effects stem from the lens as well?

 

[Robert Jackson]

shutter or not, could these effects stem from the lens as well?

Nope.

 

[Jason Rodriguez]

I just wonder the reason they designed the sensor that way. A global shutter is undoable at this size and or pixel count or what’s the reason?

A FIT or IT CCD which we're all used to, has the read-out column next to the pixel, so it's only a one-pixel transfer to a light-blocked "storage" area, and the charge can be transfered globally to this pixel storage area, allowing the pixel to reset and start integration again on a global scale. While integration is happening on the now globally reset active pixels, the storage areas are emptied and read-out. So basically read-out and integration can happen at the same time because of the storage area on a FIT or IT CCD. Other CCD architectures that do not have the storage area on-chip, like Frame-Transfer and Full-Frame require shutters to shield the imager as the pixels are read-out (or else you end up with really bad smearing).

CMOS architectures are a lot different than a FIT CCD. The most common CMOS architecture is a 3-transistor APS architecture (3T), and there is no capacity in this design for a "storage pool" like an IT or FIT CCD has. Therefore a pixel value cannot be integrating and read out at the same time. Instead, in a 3T design, a row must be read out, reset, and then it can begin integrating again. This happens on a row-by-row basis all the way down the chip, hence the "rolling shutter". The rolling shutter "skew" results because the top and bottom line of the chip are starting integration on a given frame too far apart. For instance, if the last row of the chip is reset and begins integrating 1/60th of a second after the first row of the chip, then there is not a lot of skew, because objects at the top of the frame are only going to be offset by 1/60th of a second. If it takes 1/24th of a second to scan from top to bottom of the chip, then the skew will be much greater because any moving objects at the top of the frame are offset 1/24th of a second in time from objects at the bottom of the frame . . . hence the "skew".

Global shutters are possible with CMOS designs, but they require more transistors per pixel. For instance, Fill Factory and Micron make 4T and 6T pixel designs which are commonly used in those high-speed cameras. The extra transistors allow for a temporary storage pool so that the pixel can offload it's charge and begin integrating again, while the storage pool is read-out . . . similar to the FIT CCD. The problem with 6T designs is that they take up more area on chip (transistors have size), so for a given pixel size your sensitivity is reduced as light senstive area is reduced and taken up by more transistors, and every transistor in a pixel adds noise and fixed pattern noise to the image . . . so you get reduced dynamic range. So in the end it could theoretically be done, but not at the same image quality.

Now of course more transistors does not necessarily mean reduced image quality . . . i.e., a 4T pixel is not necessarily worse than a 3T pixel depending on what you are using the extra transistor to-do. But in general, increasing the number of transistors on a pixel (and for a global shutter you typically need around 5-6 transistors) adds noise (especially fixed pattern noise), and reduces dynamic range, unless you enlarge the pixel allowing for a larger well capacity which increase the amount of electrons the pixel can gather . . . if you look at some of the global shutter sensors on the market, you'll see they have pixel sizes at 12um and larger.

 

[Gavin Greenwalt]

Lots of "yeah but's.." like you have to start a pan so you get the plates/frames .. need to film more to one side to begin with etc... but just a thought. There's always options...

Yeah but... you're assuming a static scene which is highly improbable. Someone walking through frame will experience the same distortion but you would need to isolate the walker. Rebuild a clean plate. Figure out exactly what the velocity of each pixel was and figure out whether the knee should have been one line higher etc and on and on and on.

I think people are just going to need to learn to live with it. Fancy correction plugins could be used if you need to correct the static elements for camera tracking (I suppose you're already rotoing out people in an auto 3d track anyway). But I doubt your finished product will be corrected for the effect.