Jump to content

Rolling shutter an issue ?


Michael Peploe

Recommended Posts

Been reading about the rolling shutter issue on Reduser. What are other peoples opinons on this. Seems the fire chief has suddenly stopped his ubiqitous heh heh heh's, fallen asleep and no response on this has come.

 

If you download the clip they have offerd, its the first shot. Watch the wagon wheel.

 

 

http://www.reduser.net/forum/showthread.php?t=1883

 

http://www.reduser.net/forum/showthread.php?t=1983

Link to comment
Share on other sites

Is it an issue? lots of people watched the whole film in all it's glory at NAB and to my knowledge no-one has mentioned it. It's only since the download clip was available that it's been raised.

 

 

Been reading about the rolling shutter issue on Reduser. What are other peoples opinons on this. Seems the fire chief has suddenly stopped his ubiqitous heh heh heh's, fallen asleep and no response on this has come.

 

If you download the clip they have offerd, its the first shot. Watch the wagon wheel.

http://www.reduser.net/forum/showthread.php?t=1883

 

http://www.reduser.net/forum/showthread.php?t=1983

Link to comment
Share on other sites

Is it an issue? lots of people watched the whole film in all it's glory at NAB and to my knowledge no-one has mentioned it. It's only since the download clip was available that it's been raised.

 

Well apparently if you want to do vfx work it might be a serious problem.

Link to comment
Share on other sites

  • Premium Member
Is it an issue? lots of people watched the whole film in all it's glory at NAB and to my knowledge no-one has mentioned it. It's only since the download clip was available that it's been raised.

OK, I looked at it. The clip is a WWI trench scene, with part of a wheel close in the first shot, and a couple others in the BG later on, wheels static, but hand held camera. Did I see the right clip? At least on my computer I didn't see anything that looked like a shutter issue.

 

 

-- J.S.

Link to comment
Share on other sites

  • Premium Member
Is it an issue? lots of people watched the whole film in all it's glory at NAB and to my knowledge no-one has mentioned it. It's only since the download clip was available that it's been raised.

 

Hi,

 

Please change you user name to you real name, its required by this forum.

 

Thanking you in anticipation

 

Stephen (moderator)

Link to comment
Share on other sites

OK, I looked at it. The clip is a WWI trench scene, with part of a wheel close in the first shot, and a couple others in the BG later on, wheels static, but hand held camera. Did I see the right clip? At least on my computer I didn't see anything that looked like a shutter issue.

-- J.S.

In the first scene, look at the wheel on the right as the camera begins its first pan left. At the point where motion blur begins to appear on the wheel, the wheel shears bottom-right. It's easy enough to fix, for that one scene, but the question is how it might effect motion tracking, since consistent geometry is critical to accurate 3D plotting. And since it is time-related, any motion in the frame at all will effect the apparent shape of an object -- even if the camera is still. Theoretically, a ball travelling quickly towards the lens would appear vertically lopsided, the top being further from the camera than the bottom at the time it is recorded. Would it be noticeable? I doubt it. But the f/x guys aren't interested in human perception -- they want accuracy or, at least, repairable inaccuracy.

Link to comment
Share on other sites

  • Premium Member
In the first scene, look at the wheel on the right as the camera begins its first pan left. At the point where motion blur begins to appear on the wheel, the wheel shears bottom-right. It's easy enough to fix, for that one scene, but the question is how it might effect motion tracking, since consistent geometry is critical to accurate 3D plotting. And since it is time-related, any motion in the frame at all will effect the apparent shape of an object -- even if the camera is still. Theoretically, a ball travelling quickly towards the lens would appear vertically lopsided, the top being further from the camera than the bottom at the time it is recorded. Would it be noticeable? I doubt it. But the f/x guys aren't interested in human perception -- they want accuracy or, at least, repairable inaccuracy.

Can you pull a freeze frame of that and post it here?

 

 

-- J.S.

Link to comment
Share on other sites

Can you pull a freeze frame of that and post it here?

-- J.S.

post-21345-1177568973.jpg

These are taken at 5 sec. 15 and 17 frames. The cyan frame (17) screened over red frame (15), and aligned so that the dirt at the top is grey (red and cyan cancelling each other out). At the bottom, the two colors show the shift in their relationship. The consistency of the dirt at the top, all grey, establishes that the camera hasn't rolled between frames to cause the difference in the wheel's bottom. Lack of motion blur shows this effect is not limited to rapid panning motion only. It's still an amazing camera, though.

Link to comment
Share on other sites

  • 2 weeks later...
Lack of motion blur shows this effect is not limited to rapid panning motion only. It's still an amazing camera, though
.

 

Sometimes I don't log in when I visit this forum and I just saw that pic. I just assumed it was the panning shot of a sheering wagon wheel. Your still shot of just before the sheering pan is a great example of understanding the rolling shutter even without camera movement. I read that RED is working on fixing it but I thought that this was the nature of a rolling sutter and not fixable. What can be done to alleviate this problem?

Link to comment
Share on other sites

  • Premium Member
What can be done to alleviate this problem?

The crux of the problem is the amount of time it takes to read and clear the chip. It's like an old rolling slit still camera shutter. You'd have very short exposures at every point on the film, but the top was exposed significantly earlier than the bottom, which gave fast moving objects a lean instead of motion blur.

 

Hard to say how hard it'll be to fix. The issues are very deep in their chip architecture.

 

 

 

-- J.S.

Link to comment
Share on other sites

  • 3 weeks later...

Every film camera has motion blur... that is why there is a suggested "do not exceed" panning speed. Arri's is different from Panavision... the shutters enter the frame from different directions. Rolling shutters in most CMOS sensors produce more "skew" than film cameras. Some are intolerable depending on the read method. We use a unique methodology. Our read is faster than most CMOS devices and a bit slower than most film cameras, but closer to film than conventional CMOS devices. We do not think that anyone will complain about the skew in our cameras, unless you really want to magnify the difference and make it a problem. My bet is that some here will attempt to do that. That would be in contrast to the cinematographers that want to use the RED camera for what it does better than film. I have always said that there are advantages to film and advantages to RED. We don't hide from the differences. The one main differences is that film is not likely to ever get much better at anything going forward. The development curve has pretty well run its course. Digital Cinema is quite the opposite. The remaining issues are being worked on feverishly... dynamic range and read/reset. In the future, digital will always have more resolution, provide a cleaner image and will be much easier to work with. My bet is the dynamic range issue will be matched in a couple of years or less. One of the reasons that the sensor in our camera is upgrageable. Read/reset will actually swing in favor of CMOS within the next year. The read/reset will be faster than film cameras in the next generation. The question will be if it should actually be slowed down to maintain the "filmic feel". Good news is that this is easy to do.

 

Jim

Link to comment
Share on other sites

  • Premium Member
My bet is the dynamic range issue will be matched in a couple of years or less.

That might be true if your reduce your definition of dynamic range to pure numbers (from black to white), but we all know that digital just does not blow out remotely as nicely as film, but clips in a very unpleasant way. Even a 2 or 3 stops increase will not help in that aspect. The only way to achieve that would be to have such a huge dynamic range that the chip never clips, so that one can emulate any blow-out in post (if one so choses), but such a chip is still very far of, if it is ever possible at all with the current technology.

Link to comment
Share on other sites

While this skewing issue may not be a problem visually to your general audience member, it will throw a matchmove or a composite out of whack - and I dare say it'll be very tricky to integrate a CG element into this shot without correctly de-skewing the plate or skewing the element to match.

 

The question will be if it should actually be slowed down to maintain the "filmic feel". Good news is that this is easy to do.

 

Jim - Good to see you've acknowledged the issue. Will you be making the read/reset user adjustable in future releases? Or am I misreading this statement?

 

Cheers,

Will

Link to comment
Share on other sites

My bet is the dynamic range issue will be matched in a couple of years or less. One of the reasons that the sensor in our camera is upgrageable.

 

Jim

 

People have been saying that for 20 years now? My bet is more like ten if ever. It is never my intent to put you down, just think you should stop predicting stuff you don't understand, as it will come back to haunt you :ph34r:

 

Your problem is dynamic range all happens in the analog part of the chip. The CMOS part is all the microscopic switching and analog-to-digital conversion that freights the signal out of there. The silicon photocell is an excellent way to gather photons, far more efficient than photographic film, but the bastard trip is getting the tiny signals out of the chip and into REDCODE! If not for that, your camera would have much lower noise and you could close the iris down 4 stops or so to stop overload, and then simple logarithm algorithms (try saying that after drink!) would give the long-awaited "film look".

 

The only other road is to reduce the quantum phase jitter of your clock signals, but all the blood has been gotten out of that stone now. (If you figure a way to do that, the cellphone companies will always be glad to hear from you :lol:)

 

The Dalsa does better, but it's not that good, and certainly not cheap like the RED!

 

Film emulsion isn't over-efficient at catching light photons, but it has low-noise built-in logarithmic amplification that current electronics doesn't match. For every CMOS pickup transistor you have now you'd need at least ten to make an analog anti-overload for each pixel. I know how it can be done in theory done but so do all your competition and the men who make your sensor! An awful lot of transistors would needed for a 12 megapixel sensor, and analog transistors are bastardly harder to make than ones for digital purposes!

 

And no, it can't be done after the chip, because once the overloaded pixel overloads its analog to digital converter, as I have said before 11111111111 will continue to look like 1111111111 no matter what you do to it afterwards.

 

However, my congratulations for providing the upgradeable design. For most manufacturers "upgrade" = "new one!" :(

Link to comment
Share on other sites

  • Premium Member

Aha. Well, that would be Red's "how we did it for the money" dirty little secret, if they have one, I guess. It's not a new problem as people have been discussing machine vision cameras for cinematography for quite some time, and it's a problem they often have.

 

I think it's a mistake to compare it to "motion blur" or "maximum panning speed" concerns, though. Motion blur has nothing to do with the readout characteristics of the sensor. Maximum panning speed is about avoiding strobing with a relatively low 24Hz refresh rate. It's not usually about having to avoid panning too fast because if you do it, the picture goes all bendy!

 

I am actully rather surprised that sort of compromise has been necessary, though. That sensor must have an enormous number of separate A/D subsystems on it, and I'd have thought the solution would be to arrange it as a shallow number of lines per A/D, and read them all out simultaneously, dividing the maximum error per line by the number of A/Ds on the chip. Presumably other engineering concerns prevented this.

 

It is very worrying from a CG standpoint, though - which is a shame on a camera which would otherwise lend itself well to shooting plates. It is very unusual to find a cinematography camera with a rolling shutter, though - I suspect this may be the only one.

 

Phil

Link to comment
Share on other sites

I think it's a mistake to compare it to "motion blur" or "maximum panning speed" concerns, though. Motion blur has nothing to do with the readout characteristics of the sensor. Maximum panning speed is about avoiding strobing with a relatively low 24Hz refresh rate. It's not usually about having to avoid panning too fast because if you do it, the picture goes all bendy!

 

I agree Phil, I think it's important to separate out the retinal skipping issue here.

 

-Sam

Link to comment
Share on other sites

It is very worrying from a CG standpoint, though - which is a shame on a camera which would otherwise lend itself well to shooting plates. It is very unusual to find a cinematography camera with a rolling shutter, though - I suspect this may be the only one.

 

Phil

 

Not sure if you actually read my post. Ours is a solution you have not seen before. And ALL film cameras have "rolling shutters". I assume they are usable for cinematography? It amazes me the conclusions that people draw out of thin air.

 

Jim

Link to comment
Share on other sites

Aha. Well, that would be Red's "how we did it for the money" dirty little secret, if they have one, I guess. It's not a new problem as people have been discussing machine vision cameras for cinematography for quite some time, and it's a problem they often have.

 

I think it's a mistake to compare it to "motion blur" or "maximum panning speed" concerns, though. Motion blur has nothing to do with the readout characteristics of the sensor. Maximum panning speed is about avoiding strobing with a relatively low 24Hz refresh rate. It's not usually about having to avoid panning too fast because if you do it, the picture goes all bendy!

 

Phil

 

Phil,

 

I think that film cameras have had this "bendy" issue for a while...see attachment :rolleyes:

 

-bruce

 

 

Link to comment
Share on other sites

Phil,

 

I think that film cameras have had this "bendy" issue for a while...see attachment :rolleyes:

 

-bruce

 

 

Bruce,

 

Can you explain what you are trying to say with this picture apart fom it being 'bendy'. If you hold up a picture as an example please try and credit the photographer, in this case Jacques Henri Lartigue. It would also be useful to know the shutter speed, lens and camera used but without some hard research that may be an impossible task. Btw it was taken in 1912.

 

Nick

Link to comment
Share on other sites

Aha. Well, that would be Red's "how we did it for the money" dirty little secret, if they have one, I guess. It's not a new problem as people have been discussing machine vision cameras for cinematography for quite some time, and it's a problem they often have.

 

 

Phil

Who is your reply aimed at Phil? A short quote is always nice, as then we know if we should reply or not. It is sometimes hard to tell :ph34r:

 

As regarding "doing it for the money" I rather thing the answer is that there is no real secret. Twenty years ago video cameras cost a lot, in order for the huge amounts of analog electronics to be built, tested and trouble-shot. All of which kept a lot of Japanese workers off the street, expecting pay packets at the end of the week!

 

(My wife just told me that paragraph sounds like a poem :lol: )

 

The change to digital processing was very long, but I think helped greatly by the push for battery life for laptop computers, and the appearance of pocket MP3 players.

 

Once a digital processing chain was established and could be built and run off batteries, the actual assembly cost of building a camera dropped violently, but were Sony and their friends about to tell their customers that? :lol: Especially since the pictures were actually better!

 

BUT, you only need to look at what has happened with consumer video cameras! They cost hardly anything today, but their playback quality is comparable to Betacams of 20 years ago. (Actually, if they are in good shape, the Saticon camera tubes still give the best picture if enough light is available, for my money anyway). I am astounded at what schoolchildren are doing with a cheap digital mini-DV and a sheep PC with a DVD burner!

 

I am currently transferring my collection of analog Video-8 to DVD, and I can't believe how awful the pictures are as compared to ones from the Mini-DV. But that Analog-8 camera cost me a fortune!

 

And I thought IT'S CCD pictures were so much better than the ones from my old JVC GX88 vidicon!

 

So I think the answer is more that RED are selling the camera for more like what it actually costs now, than it may have back in 1994!

 

As regards the "rolling shutter" it appears that they are simply altering the purge/readout sequence for the different blocks of sensors. To me this does not sound especially difficult, if it was anticipated before the design of the chip. But unless the chip has a very large number of readout blocks, I can't see how that would really look like a film shutter.

 

To me the easiest way to make it look like a mechanical shutter is to have a mechanical shutter!

Link to comment
Share on other sites

Your problem is dynamic range all happens in the analog part of the chip. The CMOS part is all the microscopic switching and analog-to-digital conversion that freights the signal out of there. The silicon photocell is an excellent way to gather photons, far more efficient than photographic film, but the bastard trip is getting the tiny signals out of the chip and into REDCODE! If not for that, your camera would have much lower noise and you could close the iris down 4 stops or so to stop overload, and then simple logarithm algorithms (try saying that after drink!) would give the long-awaited "film look".

 

The only other road is to reduce the quantum phase jitter of your clock signals, but all the blood has been gotten out of that stone now. (If you figure a way to do that, the cellphone companies will always be glad to hear from you :lol:)

 

In some of its cameras Fuji uses a special CCD which has photodiodes with different sensitivities. The camera combines the results from the two kind of photodiodes and produce a high dynamic range image (they claim only 2 extra stops).

 

I think this is a much better way to increase dynamic range than trying to increase dynamic range of the individual photodiodes. At least after a certain point. Because now you probably want to use all your photodiodes to make the resolution of the chip as high as possible. But as the number of pixels increase on sensors, at some point you will choose to increase dynamic range rather than resolution.

Link to comment
Share on other sites

Phil,

 

I think that film cameras have had this "bendy" issue for a while...see attachment :rolleyes:

 

-bruce

Bruce,

 

Can you explain what you are trying to say with this picture apart fom it being 'bendy'. If you hold up a picture as an example please try and credit the photographer, in this case Jacques Henri Lartigue. It would also be useful to know the shutter speed, lens and camera used but without some hard research that may be an impossible task. Btw it was taken in 1912.

 

Nick

 

Hi Nick,

 

I picked the Lartigue photograph because it is always used as an example of the effect of focal plane shutter exposing movement and the associated distortion that results. Spinning shutter motion picture cameras have a similar shutter and can create the "bendys" as things move across the frame.

 

Please accept my apologies to the heirs of Mr. Lartigue as I naively thought that his photograph was so famous as to not need a credit :unsure:

 

While we're on this subject, I've often heard people remark that the motion of electronic motion picture cameras seems odd to them, even when shot at 24p. Perhaps we have grown so accustomed to the "look" of the spinning shutter from movies that images from digital shutters are perceived a little differently.

 

And, please excuse my ignorance: Do electronic imagers expose every pixel simultaneously, or do they scan from top to bottom (or side to side)?

 

-bruce

Link to comment
Share on other sites

While we're on this subject, I've often heard people remark that the motion of electronic motion picture cameras seems odd to them, even when shot at 24p. Perhaps we have grown so accustomed to the "look" of the spinning shutter from movies that images from digital shutters are perceived a little differently.

 

And, please excuse my ignorance: Do electronic imagers expose every pixel simultaneously, or do they scan from top to bottom (or side to side)?

Yup. There's still a difference in perceived motion characteristics between video and film, even when both are shooting at 24fps at a 180-degree shutter, IMO. Given the same frame rate, I just assume it's due to the differences between the way a spinning mirror exposes an area of film, as opposed to the way a digital sensor is exposed and read-out? Perhaps also adding slightly to the difference, may be the in-camera, 2:3:3:2 pull-down cadence often used in video acquisition (as opposed to PsF)? Perhaps someone more expert can elaborate.

Edited by Ralph Oshiro
Link to comment
Share on other sites

Rolling shutters in most CMOS sensors produce more "skew" than film cameras. Some are intolerable depending on the read method. We use a unique methodology. Our read is faster than most CMOS devices and a bit slower than most film cameras, but closer to film than conventional CMOS devices . . . Read/reset will actually swing in favor of CMOS within the next year. The read/reset will be faster than film cameras in the next generation. The question will be if it should actually be slowed down to maintain the "filmic feel". Good news is that this is easy to do.

Oh. I only just read that last post on this topic. Interesting stuff, Jim!

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...