"Rolling Shutter" Skew Tests and Grabs

[Bob Diaz]

I'm with Eric on this issue, 1/250 of a second with any shutter (rolling or global) would cause problems. With a rolling shutter, as it "moves down" the lights blink ON-OFF-ON-OFF, thus we'd have 2 bright bars and 2 dark bars.

With a global shutter, the overall brightness of the image would rise and fall as the camera's shutter opens and closes in and out of sync with the times the lights are bright.

A time of 1/120 of a second would be best. That way, any part of the image would be exposed for exactly 1 cycle.

IF I want to sample audio, 2x minimum would be the rule, but anything higher would always be better. However, in this case, we are not generating a waveform from our samples, so Nyquist does not apply here.

So, it's best to match the exposure time to the time of one cycle, 120 Hz or 1/120 of a second.


Sincerely,

Bob Diaz

[Raymond Schlogel]

Well the rolling shutter finally reared its ugly head. Had a shoot last night of a live band, big crowd and with it lot's of camera's. I hate to say it but there are quite a few shot's that would have otherwise looked pretty decent that are instead "wtf" moments if anyone else were to look at em. Like the police car, the whole line across the frame with have of it being bright is a lot more painfully obvious than I had hoped it would be just reading about it and not experiencing it myself. From what I'm reading there really aint a whole lot I can do about this is there? I shoot bands quite often so a wee bit concerned.

- Ray

[Leonard Levy]

This certainly sounds like the camera's weak point.
I'm interested to hear more details about your footage.

Did it screw up your whole shoot?
Was it just still photographers or di the bands lighting also set off problems.

Please let us know if you find any "quasi fix" in post, maybe just whiting out the whole image? Do the frames with opposite half exposures match up so you could blend them? (I imagine they wouldn't be perfect though.)

[Charles Dasher]

I was watching the news the other night and they were showing some paparazzi footage in the upper corner. i can not say for sure if it was from an EX1 but they kept looping a clip of some poor celebrity and the whole thing was showing over exposed bands as the flashes went off around the scene.

Makes me glad I will be shooting most of my work outside in broad daylight.

[Raymond Schlogel]

Nah, didn't screw up the whole shoot (if it did I need to find a new job!). The stage lights had no issue what-so-ever but what it did mess up a fair amount of lot of shots from the stage that were pointed toward the crowd and several that didn't have the camera/crowd at all but someone took a pic. Gotta love camera phones. Only fix I'm aware of is go to another shot. As one can expect from this cam so much of the footage is simply stunning but it's disheartening to have this one painful caveat.

Will post some screen grabs as soon as I'm able.

- Ray

[Jamie Baughman]

Raymond,

I'm heading out to Austin to shoot some SXSW parties and shows in March, so I'd love to hear your experience with shooting live bands in the coming weeks.

I'm also trying to figure out what to do with the bands/labels/production companies who don't need an edit, but just want raw footage when I'm done...but I guess that's for a different thread!

[Steven Thomas]

I know it's a pain, but have you tried fixing this in post?
Obviously, this would be crazy if there were non stop flashes.

I guess you'll have to minimize your shots directly into the audience.

[Graeme Fullick]

I use a 80-90% transparency white frame over the top to fix the flash problem. Works perfectly every time in Avid Liquid - should work just as well with other editors. I know that it would be a lot of work with constant flashes - but this is pretty rare - even at weddings.

[Eric Pascarelli]

Graeme,

Is that 80-90% transparency or 80-90% opacity?

Is is mostly white or mostly transparent? I would think you would want something mostly white to mask the partial exposure nastiness.

[Leonard Levy]

Graeme, Could you post before and afters and explain for the dummies about post (me) exactly what your talking about.

[Graeme Fullick]

Sorry Eric,

Yes that should read opacity - writing in too much of a hurry.

Leonard,

I just paste a 80-90% opacity white frame over the offending frames, and it looks on playback like a normal flash from a still camera. I have found that a photographers flash is almost always 2 frames long, so I have set up a short cut in liquid which is a 2 frame 80% opacity white frame, which I simply pull straight over the top of the offending footage. You can even set it up so that it matches the half a frame (in CMOS cameras such as my Sony A1 and the EX1) if you wish by setting it up cropped to match the offending frame (i.e. the white opaque bit only crops over the part of the frame that is not flashed - but this sometimes requires a bit more fiddling to get the intensities of the 2 halves well matched). In Liquid this takes all of 10 seconds to accomplish. I have been using it for 2 years to cover the frame crash associated with HDV when it encounters a photographers flash (an ugly square pixellation effect) - and have yet to have a single negative comment about how it looks on any of my finished work. Liquid is a native MPEG2 editor, and actually fixes these HDV artifacts if you make it re-render them with a partially transparent clip over the top. I often use a 100% transparent white frame to repair HDV codec crashed associated with extreme motion. It does cost a tiny bit of resolution, but fixes the HDV codec crash artifacts extremely well.

If you just look at the still frames, you can often see that all of the frame is not a uniform shade of white from the flash, but at normal playback speed it is entirely imperceptible (and that is what matters), as clients almost never watch a video frame by frame!

Hope this helps.

[Brian Drysdale]

I've just been reading a CML thread about the RED's rolling shutter. It was pointed out that some CMOS cameras like the Dalsa, Phantom V and Arri D20 don't have this effect. It describes how the Dalsa has two layers, the first of which collects the light, then the light information is dumped down to the second layer (blind to light), where it's removed in packets in a rolling shutter style. The net result is that the Dalsa CMOS sensor acts like a global shutter.

It would seem that these global shutter CMOS sensors are more complex to produce, so the rolling shutter found on most CMOS cameras is a cost trade off.

[Illya Friedman]

Originally Posted by Brian Drysdale

I've just been reading a CML thread about the RED's rolling shutter. It was pointed out that some CMOS cameras like the Dalsa, Phantom V and Arri D20 don't have this effect.

The DALSA cameras actually don't use a CMOS censor. It's a CCD with a spinning mirror shutter to complement the optical viewfinder.

I.

Illya Friedman
DALSA
Digital Cinema Division
www.dalsa.com/dc

[Noah Yuan-Vogel]

sorry to resurrect this old thread, but i was playing around with a EX1 at Abel today and am buying one very soon, but while i was there i tried zooming in and whipping the camera back and forth while zoomed in. i thought we had concluded earlier in this thread that the cmos pixel readout rate is constant no matter the format (1080/24p 720/60p etc), but in my quick test i noticed that there was a lot of skew present when shooting at 1080/24p HQ, but skew effects were almost unnoticeable when shooting at 720/24p overcranked to 60p. i was viewing this on a panasonic (17" i believe) HD monitor through SDI, and at 720/24p overcranked, it appeared to be outputting 720/60p over HDSDI. at 1080/24p skew was very noticeable, more so than i remember it being when testing the ex1 previously, a huge contrast to 720/60 where the image had very smooth 60fps motion and what appeared to be much much less significant skewing. were the tests wrong? any other ideas would be very much appreciated. could it have been old firmware?

i didnt get to test it that extensively, but if someone who has a camera readily available could do further testing, it would be very interesting to see. if not, i may go back and do some more careful testing and make sure i check the firmware version as well as all the other settings.

[Perrone Ford]

From the user tests I've read, setting the camera to any of it's normal rates 1080/720 24/25/05/60 all write the normal 35mbps to card. However real overcranking actually writes closer to 70mbps to the card and may burst higher.

Obviously, this affects record times, so that must be taken into account, but it was a very nice bit of information to get.