Rolling Shutter Artifacts

[Kenn Christenson]

I do understand these artifacts are essentially "native" to using this type of CMOS sensor. I am just interested to learn about the type of "fixes" the Red team may employ on the production models.

[jbeale]

There have been posts elsewhere about research into fixing this in post if necessary. In my opinion that would be more practical than fixing it in camera, unless you actually had a full-frame type shutter, which as I understand it is something fundamental to sensor design and this sensor is not designed that way.

[Kenn Christenson]

I understand that theses artifacts are the result of the rolling shutter used in this CMOS sensor. I've also read about the "software fixes." I'm primarily interested in learning if the Red team will or can "fix" these artifacts.

[Gavin Greenwalt]

The only way to fix them would be to install a gyro and accelerometer to detect pans and compensate. That's not going to happen.

[Kenn Christenson]

The only way to fix them would be to install a gyro and accelerometer to detect pans and compensate. That's not going to happen.

I agree with you, that "fix" definitely won't happen. I was thinking along these lines - http://findarticles.com/p/articles/mi_qa3957/is_200011/ai_n8903691

[Zk2007]

If they knew a rolling shutter would be problem, any reason why they didn't design the sensor as a global shutter? I mean they did design the sensor from the ground up right? It seems to me it's counter productive to have to work on 80% of your footage to correct this. Specially in 4k, how much time and inconvenience will this add?

[Álex Montoya]

well, I don't believe that the effect is worrying. Only if you needed to add 3d would you have some trouble.

The workflow for addin cg to the image would be compensate, track, add cg and decompensate the composited plate again

[WesVasher]

I noticed the rolling shutter artifacts in the Crossing the Line clip only because my HV20 does the same thing. It's hard to see unless you are frame by framing footage though and I can't imagine anyone not looking for it to notice it.

[Elizabeth Lowrey]

The footage was gorgeous in almost all respects. Tremendous clarity, latitude, gamma distribution, obviously the DOF of 35mm. Truly, truly exciting.

But I noticed right off the bat some almost interlaced-looking artifacts at some places where there was a lot of subject movement, i.e., the frame seemed to momentarily flash visible horizontal lines of content rather than a single, unified frame. These artifacts morphed into regular motion blur if you stopped playback at that instant and viewed the frame as a still rather than in the context of playback.

Is this the "rolling shutter" you're talking about?

[Kenn Christenson]

The rolling shutter effect can be seen most dramatically in the very first shot as the camera pans to the left, watch the wheel on the right side of the frame.

[Bruce Allen]

To me, rolling shutter is part of the whole price-performance "sweet spot" for digital cinema cameras right now.

In the extreme motion situations where you see rolling shutter, you probably would be tweaking by hand anyway because it's difficult to get an accurate track then.

Wes, I am interested - do you get more rolling shutter in 24fps vs 60i mode on your HV20?

I'm just wondering if (for those EXTREME and EXTREMELY RARE situations where it is a problem) we shot 48fps (or 120fps?) on the Red and sampled every second frame on the Red if it would make rolling shutter problems less... you'd have to add motion blur in post (which sucks sometimes) but then if you are doing a lot of CG, you won't want too much motion blur in your filmed image anyway.

I think that as long as people aren't shooting crazy test material with strobes, etc, they are fine. Some numbskull will probably manage to shoot a weird test that "proves" that it is a problem, but we filmmakers won't care. Hope Red opponents don't use it to scare off technically-challenged producers though.

Anyone shot a wedding with an HV20 or Sony A1U (which also has rolling shutter) yet? I'm interested in how still camera flashes show up.

Bruce Allen
www.boacinema.com

[Thom Steinhoff]

To my eyes it looks amazing and as long as I know about a little issue like this, I can plan around it. I think constructive conversation around the topic is good, but we shouldn't take anything away from the quantum leap that the Red team has achieved. Nobody is going to see this as the motion is running, you'll only see it if you step frame by frame.

I did do a bit of poking around to learn more about it and found this solution which also illustrates the problem:

http://download.micron.com/pdf/articles/truesnaparticle.pdf

Says their technology can work up to 500fps and it does it by snapshot buffering the image then scanning the buffer instead of directly scanning the sensor.

All I know is that Red is light years ahead of anything out there and with their upgrade ability, as new sensor packs come out that enhance the abilities of the sensor technology then I can upgrade and get the latest technology they have to offer ensuring the camera I use will always be state-of-the-art.

But for now, the biggest limiter on the quality I can get off of the Red will be my personal abilities as a film maker--a warp on a small corner of a single frame when I'm shaking the camera is the least of my worries.

[garageman]

Such anomalies aren't entirely troublesome and can become "characteristic" of certain tools. It's not something that's hugely distracting. Think of the anomalies on old anamorphic lenses, people have come to expect them in some cases.

[Jance Allen]

This is my biggest concern at the moment - tracking and compositing 3d elements to Red footage. The rolling shutter shear could be really problematic, while the rest of the Red advantages for post are so superior - less or no grain, no gate weave, 4:4:4 or close, the /i data, etc. I'll be curious to see how this shakes out and if the Red team is considering this issue. Otherwise I've been blown away by their accomlishments.

[Jeff Kilgroe]

The rolling shutter effect can be seen most dramatically in the very first shot as the camera pans to the left, watch the wheel on the right side of the frame.

I saw it first thing too. But I'm not going to stress over it, or at least not yet. Lots more footage and tests to come with more enabled cameras. We'll see what happens. There's plenty of high-end cameras on the market with scanline sensors or "rolling shutters" and it's hardly the first thing that people complain about.

[WesVasher]

Wes, I am interested - do you get more rolling shutter in 24fps vs 60i mode on your HV20?

Bruce, that's a good question and it's something I've been meaning to test. I know for a fact that rolling shutter shows up at both 24p and 60i, though I don't know what is more apparent if either. I'll have to do some tests at different shutter speeds in both modes to find out.

I've tried to matchmove some hand held footage from it and it was challenging.

[Thom Steinhoff]

But I noticed right off the bat some almost interlaced-looking artifacts at some places where there was a lot of subject movement, i.e., the frame seemed to momentarily flash visible horizontal lines of content rather than a single, unified frame. These artifacts morphed into regular motion blur if you stopped playback at that instant and viewed the frame as a still rather than in the context of playback.

Is this the "rolling shutter" you're talking about?

I think this is just your PC's ability to playback footage of this size. On my PC, I see what you are talking about--but on my Mac it looks fine.

The Rolling shutter artifact is seen mostly on a small warp of the wagon wheel in the beginning of the piece. It is very subtle and can only really be seen (to my eyes) when you step frame by frame and even then it is just a couple frames that show a slight warp.

[Elizabeth Lowrey]

I don't notice anything odd with the wheel. The camera move is (intentionally, I'm sure) not the smoothest, with a little jog/shake in addition to the pan. But I don't notice any problems with the wheel.

But is anyone else seeing the fleeting horizontal lines I'm talking about? They resemble the kind of lines you'd see if you did a frame capture of interlaced footage where there was a lot of very fast motion and you did not do field blending or discard one of the two fields. I'm not sure if it is something in the footage or something in the playback of the quicktime.

I'm using a pretty new Gateway FPD2185W widescreen 21" LCD monitor with my computer. Don't think the problem is with it, although I'm not 100% certain.

[Jeff Kilgroe]

I've got nasty shearing with my Sony HVR-HC3 cmacorder. Actually the HC3 is a complete piece of junk, but that's something for another forum. Obviously, the effect is far worse with slower shutter settings... Which brings me to my concern for RED and the rolling shutter. If the shutter angle is opened up for longer frame exposure times, what will we see then?

I'm eagerly awaiting more test footage.

[Elizabeth Lowrey]

I think this is just your PC's ability to playback footage of this size. On my PC, I see what you are talking about--but on my Mac it looks fine.

The Rolling shutter artifact is seen mostly on a small warp of the wagon wheel in the beginning of the piece. It is very subtle and can only really be seen (to my eyes) when you step frame by frame and even then it is just a couple frames that show a slight warp.

Thanks, Thom. We crossed posts.

I saw the "warp" of the wagon wheel but just assumed that the camera move, which wasn't smooth, slightly dipped forward as it panned, thus making the wagon wheel appear slightly bigger for an instant. Are you guys certain that that isn't a factor?

[garageman]

When I pulled the clip into my Avid and played it back in the timeline monitor I could see the horizontal lines at certain cut points but when I play it full screen it's fine, so I think it's maybe your monitor or processor.

[Jance Allen]

I agree that the wagon wheel shutter warp isn't hugely distracting when watching the movie. But that kind of artifact can prevent a matchmove solve, and it can stand out next to composited 3d elements that don't have the artifact. It's a similar challenge to lens distortion, except there are some fairly straightforward workflows for removing/adding optics compensation in that case. I don't relish the thought of hand-warping elements to match the shutter shear. I'm waiting to see and learn more from Red.

[Jeff Kilgroe]

Thanks, Thom. We crossed posts.

I saw the "warp" of the wagon wheel but just assumed that the camera move, which wasn't smooth, slightly dipped forward as it panned, thus making the wagon wheel appear slightly bigger for an instant. Are you guys certain that that isn't a factor?

I caught it out of the corner of my eye on my second viewing full-screen (30" display). So I looked at it on a frame by frame basis. Yes, it's a shearing effect, and it's ever so slight -- not sure how I noticed it. If this is as bad as it gets, then there's nothing for anyone to worry about. I'm not concerned, I have faith in the RED team and their product. For now I just want to see more footage, especially with different shutter settings and lighting setups, various ranges and speeds of pans or subject motion. Lots to test... As as been said so many times, this was shot on alpha revision cameras. No reason to get any feathers ruffled up yet.

[Álex Montoya]

Are you sure the shearing is not combined with the unsteady camera?

I mean, I believe that a good part of the appearent distortion is just a slight change of perspective of the camera when panning.

[Steve Gibby]

Very shortly, in the field testing of RED #8, I'll shoot some fast motion kiteboarding in 2k REDCODE RAW @ 60fps. I'll overcrank and undercrank. I suspect that if there are any rolling shutter artifacts, they are accentuated by fast pans and fast lateral motion at low frame rates - like the 24fps that the PJ footage was shot at. Remember, the PJ project was test footage on prototype cameras. I viewed the PJ footage in 4k at RED and at NAB, and to my experienced eye I saw very little distracting artifacts, and to the uneducated audience's eyes, they probably wouldn't have noticed anything out of the ordinary. As pros we're trained and paid to closely analyze footage, and rightly so, but at this point RED knows what the final tweaks of the RED One camera need to be. Before anyone jumps to a conclusion that rolling shutter artifacting is going to be a significant problem with RED One, they should just wait until the cameras ship, and are tested, like my #8 test shortly, analyze the results, and then draw some preliminary conclusions. Even then, nothing is static, everything is dynamic. IMO RED has listened all along and certainly will continue to do so - and refine their products.

IMO the PJ piece was very well done under very limiting and trying circumstances. I think many of us techno-geeks simply aren't comfortable unless we have something to complain about! It gets to be where anyone who has anything positive to say about a product is labeled as a fanboy - while the hallmark of being "genuine critics" is to be negative about almost everything. When I analyze technology, it's like viewing a rose: I see the thorns and petals simultaneously. In fact how do you even know what a petal is without knowing what a thorn is? That said, while acknowledging any thorns, I tend to concentrate on the petals.

All the way through this development process of RED One, RED has solicited and analyzed end user input. IMO that won't ever stop. It's the personality of the company, and it comes down from the very top - from Jim. IMO they'll finish off this development process thoroughly and effectively.

[Thom Steinhoff]

Jeff,

I agree, no reason to worry at all--for me it took a few minutes even to see what everyone was talking about--it's the least of my worries.

Did you check out this technology--they claim to provide a global shutter solution that's cheap enough to even use on cell phone and for industrial cameras at up to 500fps.

http://download.micron.com/pdf/articles/truesnaparticle.pdf

[Jance Allen]

Yes, I'd like to clarify that I feel very positive about Red and the footage. I came away from NAB feeling like this was a watershed year for filmmakers, with Red, Redcode, Final Cut integration, Apple's Color and Stu's Looks. I don't want to dampen anyone's enthusiasm with my comments. It's just that I come from the post/vfx side of things where I do spend a lot of time stepping through frame by frame at 400% zoom, and a typical task would be to put a 3d wagon wheel next to that live action wheel and it has to look perfect.

[Elizabeth Lowrey]

Okay, I had been viewing the file only in Quicktime player.

This time I brought it into Vegas 7 and set up a project with the same size and properties as the file. Voila, the "lines" I was seeing in quicktime playback disappeared completely.

I was also able to go frame by frame and noticed the warp of the wheel. The box the guy's holding to his face also warps a bit just as the pan starts, and then everything seems to get back on track.

I can see how this would be of concern in some instances, but it wasn't noticable to me until I took it into Vegas and viewed it frame by frame. I'm just relieved that the lines I was seeing were due to some issue with my Quicktime and weren't inherent in the footage.

Is the shear effect only evident if panning too quickly? What are the variables that make it appear?

[vanguy]

I know from my days shooting with an SLR with a focal plane shutter at higher speeds, that this kind of "rolling shutter" effect could be seen on moving subjects.

Now, I'm speculating, but I'd think a film camera with a rotary shutter (especially with a narrow shutter angle) would do something like the same thing, wouldn't it? And what's more the effect would be radial, with an axis at one side of the frame.

Does anyone know more about anything like this?

[Jeff Kilgroe]

Are you sure the shearing is not combined with the unsteady camera?

It's there, but camera motions could definitely have enhanced the effect of shearing. I'm also wondering if we're seeing it more than we should because of our displays. ...Or at least this would apply to anyone on an LCD display or other monitor that doesn't refresh at an even multiple of 24hz. In other words, pulling the 24p up to 60p could be augmenting the effect by introducing judder.

Stepping though the video one frame at a time shows that the effect is very subtle and only visible on that right side of the frame for some reason. Other lines like the soldier's M1 or the periscope don't appear to be affected and they should be. So I'm a bit puzzled about that. Or as I scrub over it a few more times, I'm willing to bet we're seeing more lens distortion than rolling shutter. And I'm starting to wonder if we're picking at this footage a bit too much and we're seeing things that are really not there or issues caused by forces beyond the camera itself.

To say it again... More tests needed.

[Jeff Kilgroe]

Now, I'm speculating, but I'd think a film camera with a rotary shutter (especially with a narrow shutter angle) would do something like the same thing, wouldn't it? And what's more the effect would be radial, with an axis at one side of the frame.

Yes, film cameras can exhibit this type of artifact as well.. Depending on the shutter implementation and other factors such as speed/angle, frame rate, motion, etc.. Just one more reason not to get all worked up yet.

[Bruce Allen]

I agree that the wagon wheel shutter warp isn't hugely distracting when watching the movie. But that kind of artifact can prevent a matchmove solve, and it can stand out next to composited 3d elements that don't have the artifact. It's a similar challenge to lens distortion, except there are some fairly straightforward workflows for removing/adding optics compensation in that case. I don't relish the thought of hand-warping elements to match the shutter shear. I'm waiting to see and learn more from Red.

I think that in the cases where the matchmove would work, it won't be noticeable. I think (hope) that the shear will only crop up when the movie is too insane for a matchmove to work anyway,

Gibby, look forward to your tests!

Bruce Allen
www.boacinema.com

[Michael Struthers]

IMO, the live RED footage was certainly no worse than 35mm film during hard pans and movement.

The only "gotchas" (and I was REALLY looking)were some noise in the black on a cockpit shot and a bit of something I couldn't identfy in the clouds shot.

If "Pieces of Eight" can find an audience with possibly the worst look ever foisted on a screen, then "artifacts" on your full-frame 35 digital camera will be the least of your worries.

I felt it was miles better than a Sony f950 picture, that's for sure.

[Jeff Kilgroe]

I felt it was miles better than a Sony f950 picture, that's for sure.

Most definitely. Even though I missed NAB and the 4K screening, the little bit I have seen here tells a lot about what the camera can do.

Anyway, to flog the dead horse one more time... I mentioned lens distortion in a previous post as a possible explanation for the rolling shutter effect.

<edit> OK, I'm really starting to like the lens distortion theory. The wagon wheel warps, the soldier's head warps, the periscope not hardly at all and the dirt/rocks beyond that don't appear to be affected one bit. Or at least that's what I'm seeing as I look at again... And I think I'm done with my over-analysis that was nowhere near scientific.

[Bruce Allen]

I agree, definitely combination of lens distortion with the rolling shutter. I also agree, doesn't look like a big problem! But I'm definitely going to do a test matchmove on a moderate-to-high-motion shot first.

Bruce Allen
www.boacinema.com

[Kenn Christenson]

Actually, the reason we're seeing more distortion on the right is because the camera is panning to the left. The imager scans top to bottom, so the top is the first to be read and the bottom - the last. So when doing a quick pan to the left - the image will "slant" to the right - vice-versa panning to the right.

[Anders Holck]

There was a longer discussion about this last year on DVinfo.com http://dvinfo.net/conf/showthread.php?t=76426&page=3&highlight=rolling+shutter

It's hard to judge from the short footage, if its really something to be annoyed about. But generally, yes I would prefer it to be minimized if possible.

[Pat]

Hi everybody, I'm new here :-)

I see the same problems with my cellphone camera and have played around with it quite extensively in the past. It can be used for interesting artistic effect, though I'd prefer if it could be turned off.

I ran a few measurements on the footage: between frame 5.17 and 5.18 the camera moves approx. 10 pixels to the right (in the 1K version). If we try to rectify the waggon wheel in frame 5.17 so that it matches frame 5.16, we need to move its lower part 8 pixels to the left. Since the picture height has been cropped to 75% in post, we can extrapolate a rule: for a pan-speed of n every pixels per frame, the picture becomes slanted with n pixels accross its whole height.

So, worst case scenario: whip pan across a cityscape with lots of verticals with a movement of half of the sensor per frame. While motion blur would hide a lot, your buildings would still get slanted at about 40 degrees to the left.

At least, that's my theory from looking at that footage. It would be great if Gibby could do such a test with RED #8.

Caveats:
- alpha cams, this might be fixed with a different firmware
- might be possible to fix in post without accelometer data: measure the horizontal motion blur for each frame, divide by shutter angle, multiply by 360, slant the picture by this amount, add missing info to the bottom left from next frame.
- it's a price/performance thing. The Red One is still an amazing value even with such a minor quibble.
- might be nice to have an official word from red considering this, especially concerning the "fix" Thom found.

[Gavin Greenwalt]

Pat your solution is great until there is someone running the opposite direction through frame.

[LighthouseMEdia]

Guys I would like to just reiterate what Graeme Natress has already said here on the boards and to me personally. That any issues you will see in the footage Red is fully aware of and working to correct, if they are not in fact corrected already.

[Evin Grant]

How exactly does a rotating shutter in a Film camera differ from the Rolling design of the Mysterium? I can't seem to visualize the diffrence other than the shape of the shutter.

[Anders Holck]

Evin, let me speculate on that.
As we don't know yet how the Mysterium chip is actually functioning, lets assume it's read line by line from top to bottom. This is what the skew in the short sample could look like was the case.
Again it's hard to see based on the content and texture of the picture. A pan across a flagpole, or a truck passing through a frame would surely give us more data.
But if this is indeed the case, the delay is a constant and we can observe a sharper skew on vertical lines during Pans, and vertical compression during tilt up and expansion during tilt down. Motion blur will of course soften the edges of moving objects a bit but the main delay is differentiated line by line.

In a Film mirror shutter like on the Arri 435 the actual shutter movement is mainly sideways, and more important because it is mounted at a 45° angle to the gate most of the shutters "shadow" will be out of focus causing the actual movement edge to be blurred significantly. This will cause some of the shuttering to become more like a general image attenuation instead of a hard edged swipe.
This is a very different effedt that an exact line by line shutter.

Again no info has been given on how the CMOS is read out so this is pure speculation based on the small artifacts seen in the short downconverted QT.

[casey warren]

How exactly does a rotating shutter in a Film camera differ from the Rolling design of the Mysterium? I can't seem to visualize the diffrence other than the shape of the shutter.

I can't visualize it either.

[Obin Olson]

I see RED is not talking much about this. I would suspect they are working on it.

It's 100% a NON issue IF and only IF you can slove motion tracking with it. otherwise its a big problem. Oh, the other thing is if you crank the shutter way up for the strobe high shutter speed look, will you see all your stuff "sway" as you pan? that could also be an issue...

RED, lets hear from the team. i'ts not a deal breaker or anything but lets hear what is REALL going on with this "issue" if you will...

[fightordie]

Could this possibly be a compression issue?

[Evin Grant]

I seriously doubt it.
E.

[Alex Boothby]

It's 100% a NON issue IF and only IF you can slove motion tracking with it. otherwise its a big problem.

I do a lot of motion tracking and I really don't think it's a massive problem.

FIRST - we are talking about shearing during whip pans and fast moving objects. Fortunately the resultant motion blur lets you get away with murder and sub-par tracks. IMHO slow, subtle camera moves are much harder to nail 100%, and here any rolling shutter shearing will not be and issue.

SECOND - this problem can be fixed by digitally un-shearing the image, similar to the way you would stabilize a shot. It would be very simple to write an expression which uses the curve of the X-pos value (from a one-point track) as a control channel for the X-shear value. The multiplication factor could be determined by a few simple pan tests at various shutter speeds, and could even be published for us all to use. I assume that the results would remain very consistent from camera to camera (sensor to sensor) with the only variables being shutter speed, and perhaps FPS (which I doubt)

THIRD - if you are dead against digitally un-shearing or stabilizing your footage (it would require a mild blowup) there is still a workaround for complex 3D and 2D tracks. First, un-shear shot as described above, then 2D/3D track this result. Now animate and render CGI element, and then apply INVERSE SHEAR to element during final comp. ILM has used a similar approach for extreme wide angle shots (like some handheld shots in Twister) - first they digitally un-warp the footage (based on reference charts shot with the lens), then they track, render, and finally re-warp CGI element for comp back onto original plate.

Note that there is nothing 3D about this shearing problem. It is a 2D phenomenon and can be solved with a 2D solution. Also it is likely to be consistent, mathematical, and not arbitrary = easy to identify and fix.

This may all sound quite involved - but it's not really. For tracking purposes, I would much rather deal with these rolling shutter artifacts than the panoply of issues that arise with film, namely film weave, excessive grain, nasty-bouncy roll-ins and roll-outs (which seem to always get selected), and worst of all - WARPING OF FILM IN THE GATE. That is a bitch to track and is a real issue with HD and 2K tracks. Red will be mercilessly free of these problems.

[Nick Shaw]

Careful use of concatenation in an app like Shake should remove the need to crop in when taking the unshear-track-composite-shear approach.

Anyway, it's not that different to compensating for lens barelling when tracking and compositing.

[Rune Hansen]

SECOND - this problem can be fixed by digitally un-shearing the image, similar to the way you would stabilize a shot. It would be very simple to write an expression which uses the curve of the X-pos value (from a one-point track) as a control channel for the X-shear value. The multiplication factor could be determined by a few simple pan tests at various shutter speeds, and could even be published for us all to use. I assume that the results would remain very consistent from camera to camera (sensor to sensor) with the only variables being shutter speed, and perhaps FPS (which I doubt)

The other day I posted some references to work being done on correcting rolling shutter artifacts and an example of the corrections. It's a bit more involved but very doable.

http://mpac.ee.ntu.edu.tw/Exhibition/rolling-shutter.php

The biggest problem is when you have opposing movements, I think. You correct one movement, in turn distorting the other.

-rune

[Nick Shaw]

Opposing movement makes things more complex, but when tracking you are generally isolating objects moving in different directions, and tracking each separately. The simple single un-shear only works perfectly for nodal pans.

[Weston Ford]

it'd be nice if the image didn't shear in the first place.

[Simon Blackledge]

Do others such as Dalsa do this?

Nick.. pain in ass for matte paintings etc.. extending building and such though, yet like in another post.. alot better than distorting neg!

[Anders Holck]

For those who can't visualize the difference, here is a approximation of the shutter movement across the frame:

Rolling shutter
http://www.holckowen.com/red/Shutter_rolling.gif

Mirror shutter
http://www.holckowen.com/red/Shutter_Mirror.gif

[Alex Boothby]

There are more involved solutions but the un-shear trick gets you very close quickly. I just tried a test on the soldier-periscope shot and it took all of 60 seconds to get it about 90% (required an an animated blowup of 2.5%). A more methodical expression derived from controlled tests would be preferable, easier, repeatable and much more accurate.

In my opinion a quick fix like this would solve the problem for 99.9999% of filmgoers, and would handily deal with 2D / 3D tracking concerns (due in part to the obscuring qualities of motion blur). I might be worried when I have to track a handheld sequence shot in a jail cell (or something with distinct vertical lines) with continuous, nauseating back-and-forth whip pans... but until then....

To summarize: when tracking, I don't like film weave, I don't like heavy grain, I don't like heat haze, and I HATE film warping in the gate. This... I can handle.

I guess I'm just happy that with Red I can do a simple split screen comp without having to stabilize both plates! :biggrin: :biggrin:

[Andrew M.]

Above article is from 2005 and since then we have major progress in the CMOS technology.
Judging on the 120fps speed of the Misterium and some CMOS technologies going to 1000fps it mean that we can read CMOS in .001 sec. I don’t know how fast is Misterium but let’s say we can read it in .01 sec.
Now we can read it line by line or row by row or at 45deg or in total random since most modern CMOS sensors you can address like a memory matrix.

It is nice to have rolling shutter though since it emulates the film camera, we were just discussing it in another thread.

So CMOS starts its cycle from shunting all the photo-sensors to discharge all of them to zero. Then it lets it charge for maximum full frame minus the read time (~360 deg shutter) where there is no rolling just the blur is accumulated because object is moving or we are panning. Now, after reading the cell we can shunt it and let it accumulate the next charge or shunt it and keep it there. The cells that were read last will accumulate a bit more charge and we get more exposure in it, which is not good unless we will read extremely fast. If we can’t read fast then we have to allow some rolling. There is some development work done to prevent CMOS from charging once the read cycle starts but then we loose on some exposure time like we loose in the camera when shutter is off.

[Jance Allen]

Boothba, thanks for all the info. I agree with the tradeoffs - I can't count how many "locked off" shots I've had to stabilize due to film's mechanical nature, and Red looks to be so superior for vfx work in general. I'll be curious for more test footage, and if the shear is an issue, I'll want to learn more about the solutions you've described.

[Simon Blackledge]

There are more involved solutions but the un-shear trick gets you very close quickly. I just tried a test on the soldier-periscope shot and it took all of 60 seconds to get it about 90% (required an an animated blowup of 2.5%). A more methodical expression derived from controlled tests would be preferable, easier, repeatable and much more accurate.


In what ? shake? share!.... ;) lol...

[Andrew M.]

Do others such as Dalsa do this?

Nick.. pain in ass for matte paintings etc.. extending building and such though, yet like in another post.. alot better than distorting neg!

Delsa is CCD so you can transfer cell charge off the main photosite and read it off line.
See link below:
http://www.dalsa.com/dc/documents/Image_Sensor_Architecture_Whitepaper_Digital_Cinem a_00218-00_03-70.pdf
http://www.dalsa.com/markets/ccd_vs_cmos.asp

[Nick Shaw]

In what ? shake? share!.... ;) lol...

In Shake you could write a DisplaceX function derived from tracking data. Not easy, but hey, it's matching those little anomalies in the live action plates that makes the difference between good compositing and seamless compositing!

[Jim Arthurs]

Too bad the sensor wasn't designed, rotated 90degrees sideways, so the read off would be in the same direction as a film camera's shutter pass-over... I mean, why would it HAVE to read top to bottom... why not design it to read left to right by simply rotating a tall rectangular chip? I'd MUCH rather have sheer happening in that fashion.

Heck, I'd rather not have sheer in the first place. I'd sure like a geometrically stable image in all conditions. I'm not in a panic over it, let's see how the actual camera's perform and not get overly anxious about the beta chips and circuitry...

[Alex Boothby]

In what ? shake? share!.... ;) lol...

In flame... ha ha... perhaps you're familiar? :biggrin:

I'm away from the suite right now but the process basically went like this:

1.) - one point track scene
2.) - copy X-pos data to X-shear of a new axis.
3.) - Jitter by 2 (because the rolling shutter artifact only cares about the subtle differences from one frame to the next, not the huge X-pos curve generated by the track )
4.) - 'YScale' X-shear curve unitl it looked right.
... and lastly...
5.) - slip entire curve by neg 1 frames. This is because the rolling shutter event anticipated the camera pan by one frame (or more likely a half frame.)

All very unscientific... but this is what I was basically getting:

- from one frame to the next, every X-pos increase of 20 pixels (from the original track) required an un-shear value of 2. So about a 10 to 1 ratio - kinda...

Now this ratio only works with a 1K file. If it was 4K the values would be 80 pixels X-pos : 2 units X-shear (or a 40 to 1 ratio)

I feel like I'm talking Klingon here.... Anyway the quick fix worked to about 90%.

It we shot some controlled tests, panning the camera left and right on a resolution chart or something, I'm sure we could build a more precise expression to speed things up.

[Nick Shaw]

This would work to 'clean' the background of a pan for tracking, which I assume is what you are talking about. After tracking, you would apply a negative version of the shear to any element to be composited with the background, then composite it with the original (un-sheared) background. Do I have that right?

It could not be used to 'clean' a shot by un-shearing and then using that in the finished version. Usually a pan would be following an actor across a scene, and the camera shear would only be on the background, not the actor, as they would stay static in the frame. Applying an un-shear would clean the background and shear the actor. This would be bad!

[Antoine Fabi]

Too bad the sensor wasn't designed, rotated 90degrees sideways, so the read off would be in the same direction as a film camera's shutter pass-over... I mean, why would it HAVE to read top to bottom... why not design it to read left to right by simply rotating a tall rectangular chip? I'd MUCH rather have sheer happening in that fashion.

Heck, I'd rather not have sheer in the first place. I'd sure like a geometrically stable image in all conditions. I'm not in a panic over it, let's see how the actual camera's perform and not get overly anxious about the beta chips and circuitry...

Ah !

Jim, you beat me!!!

I was going to to make a joke of it...

...but it would be the easiest way.

Me too, not anxious at all.

[Alex Boothby]

This would work to 'clean' the background of a pan for tracking, which I assume is what you are talking about. After tracking, you would apply a negative version of the shear to any element to be composited with the background, then composite it with the original (un-sheared) background. Do I have that right? !

Yup exactly. If it was me I would just un-shear the plate and make that the new BG (like a smoothcam render) - but this comes at the expense of a blowup (or something more clever)



Usually a pan would be following an actor across a scene, and the camera shear would only be on the background, not the actor, as they would stay static in the frame. Applying an un-shear would clean the background and shear the actor. This would be bad!

When you're right, you're right! :biggrin:

But humans are organic. Motion picture footage is organic. These things are often forgiving. I'd be concerned if it was some really hard-edge environment like the wall-o-monitors scene in the second Matrix. We'll have to wait and see I guess.

I think the PJ short is probably a great, gritty, real world example with lots of movement. I know the hard core techies spotted this issue when reviewing the posted quicktime (and some were looking - cough, cough), but I wonder if anyone noticed it at the NAB projection? (don't know - I wasn't there).

If the topic at hand is about motion tracking, where accuracy is required, I think these kind of issues are par-for-the-course in vfx land. It just doesn't strike me as an insurmountable obstacle. I mean, I've never even had two MoCo plates line up perfectly... Let alone "Human MoCo"...

[Nick Shaw]

Certainly none of this struck me when viewing at NAB, but I have to admit I was rather caught up in it as a movie (I regard that as entirely a good thing!) and found it hard to step back and apply my techie mind to the footage.

[Antoine Fabi]

I saw it twice at NAB with two friends that are extremely familiar with image quality. One guy was a film guy, and the other one is in charge of QC in a major broadcast network.

We never noticed...honestly.

We were simply charmed by what we saw. (We still are)

[Rune Hansen]

but I wonder if anyone noticed it at the NAB projection? (don't know - I wasn't there).

Watched it twice at NAB. Didn't notice the sheering -- the first time I was just too caught up in the way it all looked so very very pretty and "how the @#$ did they shoot all this in two days?" Did notice the FPN, though. Second time I spent more time looking into the images and I admit it all looked very good. To me the 1K QuickTime doesn't show the most impressive images. You ought to see the dogfight scenes, which are just mindbogglingly awesome...

:-)

-rune

[Seth Larney]

THIRD - if you are dead against digitally un-shearing or stabilizing your footage (it would require a mild blowup) there is still a workaround for complex 3D and 2D tracks. First, un-shear shot as described above, then 2D/3D track this result. Now animate and render CGI element, and then apply INVERSE SHEAR to element during final comp. ILM has used a similar approach for extreme wide angle shots (like some handheld shots in Twister) - first they digitally un-warp the footage (based on reference charts shot with the lens), then they track, render, and finally re-warp CGI element for comp back onto original plate.
.

Boothba is spot on with this. In fact, as alot of the guys here will know, lens distortion unwarping is done everyday on studio fx pictures (more than you may think, on Superman for instance, we did it because Maya doesn't handle lens distortion), it is extremely commonplace. Sure dealing the shear will introduce another step in your workflow for tracking certain shots, but realistically, the number of shots you will need to do this on should be pretty minimal, and once you have an efficent workflow in place for dealing with this it won't be such a big deal (ala lens unwarping). Also, as has been mentioned, there is always the posibility of hand tracking over the affected areas of the shot.

Cheers,
Seth.

[McDiver]

I don't think fixing the sheer is as easy as fixing lens distortion. Lens distortion, once corrected for a shot, applies to the shot regardless of the apparent motion of the objects involved. It's a constant (assuming there's no zooming going on).
A rolling-shutter sheer, however, is dependent on the (apparent) motion of each object within a scene. The more it moves during a shutter's top-to-bottom capture, the more the bottom portion of that object shifts in relation to the top. How much it shifts is dependent on the scanning speed of the chip and the apparent motion of the object.
With a perfect nodal pan, one could expect all elements to sheer equally, since all elements cross the image plane at the same rate of speed. So, for those rare shots, a simple 2D track could be used to automatically correct the sheer that appears equally throughout the shot. Those rare shots, however, are never used to extract 3D camera motion -- since there is none ;) And if the viewer isn't going to notice it, then maybe there's really no point.
For those trying to get a handle on the real problem, imagine a low-angle dolly down a fence-lined road, with distant telephone poles reaching into the sky. The fence posts, say 6 feet from camera, will travel across the frame much faster than the poles 100 feet away. Therefore, the fenceposts will sheer at a greater angle than the distant telephone poles would, even though they share the same plate and the same height in the image plane. If you then correct the sheer for the poles, the posts stay somewhat slanted. And if you correct for the posts, the poles end up sheering the opposite way. And this totally ignores objects with their own motion, to boot.
So, to extract the motion of a camera travelling down that road, you'd need to somehow correct for each object you plan to track, and do it accurately enough to base your 3D composite on.
Otherwise, one could expect the 3D tracking software is going to have trouble getting a solution -- every time the camera moves, the 3D data will plot as if all points low in the scene start moving closer to the camera (thus mathematically explaining the exagerated motion at the bottom of a scene.) It would be very confusing.

My thoughts on Red, since this is my first post: Awesome! Can't wait to get one! I'll never look back! If sheering turns out to be its only "problem," I'll be celebrating for years. I still can't believe this is happening! . . .

[Tony Lorentzen]

I'm wondering why nobody at RED has commented on this issue at all. Isn't this a serious issue with the RED?

[Jannard]

Take another swing at analyzing it when we have a finished camera. You are looking at footage shot with Alpha prototypes. Lots has changed from then to now. Shear is an issue with every film and video camera. It is one of a hundred issues we are still working on.

Jim

[Tony Lorentzen]

Thanks Jim - nice to know that this is an issue you're looking into. I was just under the impression that this was so deeply integrated into the sensor - thus making it a non-fixable problem at this time in development. I'm sure people will put the first "real" cameras through a great ordeal of tests.

[Jannard]

I was just under the impression that this was so deeply integrated into the sensor - thus making it a non-fixable problem at this time in development.

I'm not sure what gave you that impression... no sensor is designed like ours. It is embedded with flexibility. One of our most difficult tasks is getting our arms around all the possibilities.

Jim

[Ace]

I wonder if a simple latency compensation algorithm can resolve this issue.. sort of like a general timecode for each horizontal line.. buffered and synced as each frame goes through the frame buffer.

[Zk2007]

As I understand it now, short of replacing the sensor with a global shutter one, there's not much to be worked on. But let's hope.

[Ace]

If thats the case (that its unresolvable), it will be an intrinsic aesthetic of the medium which isnt a totally bad thing. People like film for its quirks.

[Brian D. Goff]

... no sensor is designed like ours. It is embedded with flexibility. One of our most difficult tasks is getting our arms around all the possibilities.

Jim

shear or not - you are putting the most exciting piece of hard ware into our hands with absolutely breathtaking quality, I can only say thank you!

[Jannard]

As I understand it now, short of replacing the sensor with a global shutter one, there's not much to be worked on. But let's hope.

Again... I'm not sure where you are getting your info but it is inaccurate as it relates to Mysterium.

Jim

[Nick Shaw]

Jason Rodriguez's post (http://www.reduser.net/forum/showthread.php?p=35514#post35514) implies it is the case. But I assume he is not privy to the detailed technical spec of the Mysterium, so he might be wrong.

[Jannard]

Jason has not seen the design of Mysterium...

Jim

[Tony Lorentzen]

Again... I'm not sure where you are getting your info but it is inaccurate as it relates to Mysterium.

That's why you're here to enlighten us, Jim :blush:

I guess we're all jumping to (possible inaccurate) conclusions because we've been told the Mysterium sensor is a CMOS sensor and these types of sensors are known to have the rolling shutter problem.

[Seth Larney]

I don't think fixing the sheer is as easy as fixing lens distortion. Lens distortion, once corrected for a shot, applies to the shot regardless of the apparent motion of the objects involved. It's a constant (assuming there's no zooming going on).
A rolling-shutter sheer, however, is dependent on the (apparent) motion of each object within a scene. The more it moves during a shutter's top-to-bottom capture, the more the bottom portion of that object shifts in relation to the top. How much it shifts is dependent on the scanning speed of the chip and the apparent motion of the object.
With a perfect nodal pan, one could expect all elements to sheer equally, since all elements cross the image plane at the same rate of speed. So, for those rare shots, a simple 2D track could be used to automatically correct the sheer that appears equally throughout the shot. Those rare shots, however, are never used to extract 3D camera motion -- since there is none ;) And if the viewer isn't going to notice it, then maybe there's really no point.
For those trying to get a handle on the real problem, imagine a low-angle dolly down a fence-lined road, with distant telephone poles reaching into the sky. The fence posts, say 6 feet from camera, will travel across the frame much faster than the poles 100 feet away. Therefore, the fenceposts will sheer at a greater angle than the distant telephone poles would, even though they share the same plate and the same height in the image plane. If you then correct the sheer for the poles, the posts stay somewhat slanted. And if you correct for the posts, the poles end up sheering the opposite way. And this totally ignores objects with their own motion, to boot.
So, to extract the motion of a camera travelling down that road, you'd need to somehow correct for each object you plan to track, and do it accurately enough to base your 3D composite on.
Otherwise, one could expect the 3D tracking software is going to have trouble getting a solution -- every time the camera moves, the 3D data will plot as if all points low in the scene start moving closer to the camera (thus mathematically explaining the exagerated motion at the bottom of a scene.) It would be very confusing.

My thoughts on Red, since this is my first post: Awesome! Can't wait to get one! I'll never look back! If sheering turns out to be its only "problem," I'll be celebrating for years. I still can't believe this is happening! . . .

McDriver, you are absolutely right. I guess in the cases where you have lots of depth to your shot (ie, the example you are talking about), then hand tracking will be the easiest (only?) solution (providing your 3D track can't be tweaked to accomodate). I still think though, that the number of shots in which this happens will be not too large, and there will always be a work around. I have tracked dozens of shots (or more) from digital chips that have shearing, and have most of the time found other elements (motion blur etc) to be more of a pain.

Having said that, I think only further testing is going to give us a definitive on this (I don't know enough about sensor tech to speculate on whether the Red boys can work around this or not) and like you, McDiver, if this is the only major concern with the mysterium (an I'm not convinced yet that it is a major concern), than I'll be pretty stoked.

[Jason Rodriguez]

Jason has not seen the design of Mysterium...

Nope, I haven't . . . but I was just outlining the general features of 3T pixel CMOS sensors and why they don't have global shutters . . . and what it does take to get a global shutter on a CMOS sensor (more transistors to act as a storage point so that integration of a new frame and read-out of the last frame's charge can happen at the same time).

Sorry to jump into this discussion . . .

[Kenn Christenson]

Take another swing at analyzing it when we have a finished camera. You are looking at footage shot with Alpha prototypes. Lots has changed from then to now. Shear is an issue with every film and video camera. It is one of a hundred issues we are still working on.

Jim

As the guy who started this thread, let me say: that's all I need to know (and October can't get here soon enough!)

[liangck]

The other day I posted some references to work being done on correcting rolling shutter artifacts and an example of the corrections. It's a bit more involved but very doable.

http://mpac.ee.ntu.edu.tw/Exhibition/rolling-shutter.php

The biggest problem is when you have opposing movements, I think. You correct one movement, in turn distorting the other.

-rune

Hi

I'm one of the author of that research work. I find here due to the massive re-direction from here. It's interesting to see a research work being noticed by real photography experts.

To be honest, I believe (and proved) that rolling shutter can not be simply or automatically corrected. Our method can correct the case where there is only camera panning (no rotating) or there is only a moving object (if the background is static and can be subtracted). If there is a perfect tracking software to find the correspondences in the sequence, it will benefit the correction method. However, I don't think to correct complex motions in the rolling shutter camera can be done automatically.

Chia-Kai

[Andrew M.]

I do not understand the problem.
The whole CMOS sensor readout could be done in 1/1000 of the second, these days.
Maybe not completely out of the chip but at minimum to the cache on the chip.

Standard mechanical shutter doesn’t even move 1% through the frame during this time.
So even if Misterium is 3T design without global shutter we end up with 1% overexposure on the last pixel being red by the processor and zero skew with minimal motion blur that is there on a standard mechanical shutter as well.

[Jason Rodriguez]

The whole CMOS sensor readout could be done in 1/1000 of the second, these days.

That would be a VERY fast pixel clock! Remember, the pixel data has to read-out to *somewhere*


Maybe not completely out of the chip but at minimum to the cache on the chip.

That "cache" you speak of are the extra transistors beyond a 3T design used for storing the charge from the photodiode in global shapshot shutter.

So instead of having to read-out the whole row before new integration can being, with extra transistors, you can store the charge from the photodiode on the pixel, begin integration of the pixel again, and read-out while the pixel is integrating (like a FIT CCD). Because the storage transistor is on the pixel itself, the transfer to this "cache" is very fast. Without this transistor, the whole row must be read-out first, and then you have the pixel clock to deal with which is determining how fast the line time will be. So while it's easy to get a line to read-out in a 1/1000th of a second, to then have then 2000+ lines read out in the same 1/1000th of a second in a rolling-shutter fashion (which is one line, then the next line, then the next line) would require a VERY fast pixel clock.

[liangck]

I do not understand the problem.
The whole CMOS sensor readout could be done in 1/1000 of the second, these days.
Maybe not completely out of the chip but at minimum to the cache on the chip.

Standard mechanical shutter doesn’t even move 1% through the frame during this time.
So even if Misterium is 3T design without global shutter we end up with 1% overexposure on the last pixel being red by the processor and zero skew with minimal motion blur that is there on a standard mechanical shutter as well.

You are correct in many points. In our survey, this distortion is serious only when a on-chip is not available. However this is a common design in webcams and some surveillance systems. In these devices the transmission rate from the device to the computer limits the readout speed.

And another point from Jason is also correct. Sometimes you can not use fast clocks for the sensor.

[tj williams]

I can imagine a lot of fixes for this. The amazing thing is the level of concern. For years I've listened to people talk about a lot of abberations of film as being that beautiful look of film. Flicker, gate weave, gamma, knee, all caused by the chemical mechanical process of film passing thru the gate of a mechanical camera, all worshipped as the only way to shoot drama..... From the third row at NAB I gotta say, what you are talking about will not be noticed by any member of any audience. If the film is so bad that this is a concern better you had smashed your camera before embarassing yourself like that. Probably this will be fixed/changed in production cameras. If not I suggest we embrace this as being even more beautiful than gate weave.

[McDiver]

TJ, I don't think this is about "embarassing yourself." It's a camera -- a tool -- an instrument . . . whatever you want to call it. All the flaws you mention for film are flaws. If someone feels nastalgic about it, or has somehow concluded that they add to the aesthetic -- well, I don't stand with them. I think they've been confused about what makes film beautiful, and have talked themselves into something. But fine, we can let that go.

There's something I've wanted to say for a couple weeks now, so here goes (time for a breath-hold).

The technical discussions about possible shortcomings of the Red shouldn't be taken as automatically negative or combative. It's not a religion. It's just a camera. An amazing camera, but still a camera. It cannot be perfect for every job, nor can it save you (which is what KRD was getting at, if I'm not mistaken) . So, what are it's strengths? What are it's (dare I say it) weaknesses? The sooner we understand it's shortcomings, the sooner we can start figuring the work-arounds, because we are going to be using it, aren't we. C'mon. Love might be blind, but at some point we have to put this gem to work!!
I'm very interested in all the little quirks. Some have said that those who obsess over tiny details make terrible movies. I say that only those who obsess over details have the preparation and experience and options to make amazing movies. Filmmaking is like going to war. Kubrick pointed out that directing is the closest thing a civilian will come to acting as a general. The SEALs have a plaque over the door to their training facility in Coronado -- The more you sweat in peace, the less you bleed in war. Preparation, details, practice -- it's all part of the craft. If people want to take this whole process lightly, that's fine. But how often has that led to success?
Some people here are only concerned with how Red looks projected at 4K while telling a story. Fine. I think those are features that Red Alpha has already demonstrated. So there's no further discussion for those people. Go away, and wait for Red to call you (I'm joking).
Others are curioius about using it for more. What kind of color saturation and contrast can we extract? (My personal observations are that everything made public looks flat, like it was shot under a giant screen tent.) It'd be nice to see variety or have some raw frames to play with. I know we'll get our chance, but I think it's OK to be curious.
Another curiosity -- can you track this camera through 3D space, and at what speed does that become difficult (not to be confused with simple After Effects-type point tracking). With the alpha camera in PJ's film, sheering begins with the faintest motion blur. OK, we understand that none of this matters at this point because that's being addressed. But curiously, Obin addressed this potential problem a year ago (http://dvinfo.net/conf/showthread.php?t=76426&page=4&highlight=rolling+shutter), and was told it wouldn't be a problem. Who was the first to notice sheering in the "First clip from Crossing The Line?" Obin. So, now the curious are more curious. It's just a point of interest to those who use cameras for more than capturing pictures.
Sometimes it feels to me that curiosity carries a negative vibe here. It shouldn't. There's no question this is an amazing camera at an amazing price point. But it's simply impossible for it to be perfect for everything. So what is its strong suit? Interest has been piqued. Bold statements made. The curious ask questions and pool their information when direct answers are delayed. When you go on a date, you don't just stop at the date's "personality" do you? :).

Here's a quote I really like. "We have become too accustomed to the romantic image of the artist as someone creating in an ivory tower not to entertain doubts about such indefatigable attention to economic, technical or administrative questions. But in Kubrick's case one can draw no strict line between his work and this kind of super-technician's existence. Some film-makers find their inspiration in the contemplation of nature, others from the study of news items, still others in constant contact with the world at large. Kubrick's films reflect his perfectionism, his inordinate taste for technology, his fascination with diagrams and statistics, but also his fear of any flaw in a totally programmed system, of an excessive dependence on machines."(http://www.visual-memory.co.uk/amk/doc/milestones.html)

Thanks -- McDiver

[Gbabymogul]

I've read over this thread several times, and there's one thing that often gets overlooked.

The audience won't care what frame reate you shoot at, or what a global vs. rolling shutter is, or any of the geek tech talk. If they're sitting there watching a movie on the bigscreen, DVD, or watching trailers on the net they only care about one thing - not to be taken out of the story by odd motion. They can't tell you why, or how, or even what, but they will mark the movie lower on their scorecards - they only care about now, not future tech.

Consider this a laymans post compared to the rest of the tech stuff (and much more eloquent dudes), but RED should only have one goal - make the motion seemless with the picture. Make it so it is as seemless as film - I've never heard someone complain about films motion (unless it's the overused Tony Scott skip printing biatch)...

After all the RED footy, we'll have to see a lot more before people start getting their knickers in a twist, IMO. I think the RED team is listenting. They don't want people to have to use supercomputers to get the equivalent of films smooth motion in post. Otherwise this film vs. video argument wouldn't even be playing in the same ballpark...and RED definately has suited up to play in the big leagues, and we're all the better for it:biggrin::beer:

[Mr. Paul White]

I've read a lot about it. I'm fairly concerned.

Links:
http://www.dvxuser.com/V6/showpost.php?p=900634&postcount=84
http://dvinfo.net/conf/showthread.php?t=92486
http://www.dvinfo.net/conf/showthread.php?t=89865
http://www.dvxuser.com/V6/showthread.php?t=95296
http://www.hv20.com/showthread.php?t=124

[Gbabymogul]

I really don't think it's possible to extrapolate from the HV20 to RED. Not that i'm an expert, by any means, but we still don't know how the shutter is done >>> mostly guessing.

We just need to see more footage, or at least i do.

Seemless motion is integral to matching films quality. I trust the team to know that.

Maybe some RED Hawaiian surfing footy :wink:

i'll bring the bikini babes. :beer:

[Álex Montoya]

Hmmm, I had not thought about the problem of shooting a scene with light flashes with rolling shooter...

[Jim Arthurs]

My first and only post on this subject.

I am actually quite concerned, but not yet in any sort of panic. Prototype cameras, new boards coming in, etc. I will wait to see what the first production cameras are doing. Thank God for the guys #8 and up and the work they will do for us!

The footage from the PJ short is perfectly acceptable to any viewer, but IF significant rolling shutter artifacts are present, then there's a world of hurt coming for FX guys like me;

Feeding footage taken by a rolling shutter into 3D tracking software is basically the same as feeding garbage info into it... it simply won't be accurate if the skew is significant. Post "fixes" to tweak the skew are inherently flawed due to the way the skew evolves dependent on apparent camera and subject motion. You NEED geometrically stable over time images to do good work! If the skew is trivial, and is only a degree or two at a whip pan speed, then no harm-no problem. Again, if's, if's, if's...

I figure since many FX houses have the camera on early reservation, and that first Russian feature is using RED for FX work, we'll know soon enough, certainly before I need to worry out at reservation 411.

I personally need two things from the RED... fantastic matting with bluescreen/greenscreen and geometrically stable images. I would have never imagined that the second area would have been one of potential concern...

[Miltos Pilalitos]

The way the film's camera shutter works might be different but the images are still not "geometrically stable" as you put it. Think about it...

However, we still 3D-track footage coming from scanned film. I have also tracked succesfully footage from DV, Beta SP and even HDV(m2t).

If the 3D tracking softwares needed 100% correct footage to create 100% correct trackers we never would have tracked anything until now.

[Antoine Fabi]

Just curious,

the distortion appears when you move the camera horizontally, because the CMOS shutter equivalent (the scan mvt) moves from top to bottom.

So, logically, we should see the very same effect with film, if you move the camera up and down, since the shutter moves from left to right.
A horizontal line should show a slight diagonal distortion...no ?

[KJames]

I have heard many references to the 20 second clip of Cross The Line as well as speculation about how the sensor is made.

The clip is a down res'd, compressed version of the original files. And no one has said anything about the technical specs of the Mysterium itself.

I've looked back and forth at every frame and not seen any artifacts. People who attended NAB and saw the original didn't see any either. I've seen motion blur. But not artifacts.

As well image motion is very subjective. If you were to ask someone who only had seen 1920s black and white movies and then showed them a modern movie I am sure they would find a lot distracting because it takes time to adapt to new look and movement.

I am very puzzled as to why so many people are trying find fault in something so new and something that has already blown away the competition. For this very reason we are not seeing the full version of Cross The Line on the net.

[Jannard]

Pardon me from repeating myself... BUT, "Boris" and "Natasha" are Alpha prototypes. You might all relax and wait to get worked up AFTER we release a production camera. There were a ton of things we can point to from our Alpha prototypes that needed to be addressed. And they are. This thread reminds me of all the fear (last year) about the RED GUN.

Jim

[Andrew Benz]

I am not scared. I know there are people that are alot smarter than me working on it... But god bless my competition, get ready to have your arri bl4 handed to you in pieces...

Just got my confirmation for the prime lens set, thank you Kelly!

*edit*sorry if my arri comment offended anyone-but there is a reason for it. BTW, I do love Arri...

[Antoine Fabi]

I'm not scared at all.

100% calm, serene :)

[James T Mather]

This is a great example of group hysteria led by rumour and uninformed half-truths.

As Jim says - Wait for the production camera. They are aware of it.

[vovin]

Pardon me from repeating myself... BUT, "Boris" and "Natasha" are Alpha prototypes. You might all relax and wait to get worked up AFTER we release a production camera. There were a ton of things we can point to from our Alpha prototypes that needed to be addressed. And they are. This thread reminds me of all the fear (last year) about the RED GUN.

Jim

There isn't a single hint of fear from me at least, I am so friggin impressed with both your expertise and your AMAZING communication with all of us, that I am almost prepared to believe you are gods in human forms :/

[jaadgy akanni]

There isn't a single hint of fear from me at least, I am so friggin impressed with both your expertise and your AMAZING communication with all of us, that I am almost prepared to believe you are gods in human forms :/

I beg to differ-gods tend to be paternalistically condescending and arrogant, and Jim is all the opposite: humble and attentive. Hold on, let me wipe the brown stuff offf my nose...lol

[Obin Olson]

What if RED recorded more then 4k wide worth of data, then you could "slide" the rows till the shear was gone no? and still keep your 4k resolution

[Jannard]

Mr. White... right after I posted:

"Pardon me from repeating myself... BUT, "Boris" and "Natasha" are Alpha prototypes. You might all relax and wait to get worked up AFTER we release a production camera. There were a ton of things we can point to from our Alpha prototypes that needed to be addressed. And they are. This thread reminds me of all the fear (last year) about the RED GUN."

...you posted the above. What were you thinking?

Jim

[Jarred Land]

and with that statement.. time to close it down.