Sunday
Mar082009
GH1 Live MOS Skew
Sunday, March 8, 2009 at 12:33AM I hate to admit it to y'all, but I went all CSI on a crappy sample video from the pre-production Panasonic GH1 to check out its jello-factor. Verdict: skew is present, but it's minimal, and hard to see without this kind of party-pooper hoop jumping.
The original video is here (and below). Props to Slashgear for making a 720p vesion available for download.
Here's some 5D Mark II footage for (highly unscientific) comparison. I assume this has been run through some kind of auto-stablization such as one finds in Final Cut Pro and iMovie 09, which is not unlike what I did to the GH1 clip.
Reader Comments (29)
I think it was pretty evident, and while quick, those pans weren't crazy fast compared to what one might have in an action flick. I agree, not as bad as the 5DII and D90, but not too good.
But, that's a pre-pro model, so... we shall see. I expect it to be more than fine for those already used to rolling shutter affects.
I'd really like to see this with d90 and 5dII Footage to compare the amount of Skew :-)
GPSchnyder, refresh the post, I included a slightly similar 5D Mark II clip.
the skew was probably expected when panning....my concern with this, is while trying to focus on someone, ("static" handheld) you`ll get almost similar jello with d90.
What I mean by that, if you move your fingers or if you are not on a tripod, the footage will jump at the slightest movement.
5DII on my end , doesn't present that much of a problem, but then again a fair comparison will have to be made with same focal length equivalent lenses
That's a nice test Stu. It appears to be better than a HV20.
Something else I think would be worth trying would be to place all these cameras on an auto pan/tilt head such as the Bescor model set to full speed. Record a few takes of each camera being panned at the same speed and then compare the results.
What would be even more interesting is to see how the larger rolling shutter cameras such as the Red One and EX1 compare also.
Maybe a magazine such a DV could organise such a test? They could call it something like 'The big rolling shutter showdown', or 'The HD camera jello shootout'.
Don't you think every video-capable DSLR nees its own Vincent Laforet? :)
Stu,
How much a factor do you think frame size has on your test? i.e fewer pixels involved, so even if you scale might minimize the effect?
But from what you have shown, it's less than my HV30 so even if it's own slightly more at full 1080p sensor capture, it should still be workable
Isn't the amount of rolling shutter that you are going to see dependent on the shutter speed?
Inside with artificial lighting and with automatic settings is probably going to give you the slowest shutter speed and therefore worst case.
Regarding pricing and delivery of this camera, I spoke to my local supplier who contacted the NZ Panasonic rep, they are saying July and reasonably more expensive than the G1.
Stu,
Your "original video" link isn't working (results in 404 error). That being said, so far the GH1 isn't looking too bad. There is some skew but it's looking relatively minimal at this point. If they can keep it down to a 5D mk.II level (or at least measurably better than the HV20) then I'm sold. Right now it is THE camera to watch. Though I'm still bummed I can't get a body only version of the cam. I can't wait for the first "subway short"!
I'd say that looks pretty bad. The pans/movement weren't all that, and still the wobble/skew reared it's ugly head pretty blatantly.
Disappointing but not shocking. I can't ask this camera to be too perfect.
I'm more interested in 17mbps AVCHD and MJPEG recording mode's quality.
Sweet idea.
I think the image stabilization is also effecting the video. It looks like one of the more bothersome "jiggles" (panning back towards kid picture) coincides with the image stabilization re-framing.
Can't wait till it hits the shelves.
I also think that you can get rid of a lot of the woobles if you steady your recording in Production. I'll get myself a ModoSteady for 99€ and hope to have good results with it.
Hi Stu,
Not really apropos to this post, but one thing I've wanted to ask you about your comments on the 5DMkII: You (and some others) have mentioned several times that the Canon camera uses binning to extract it's 1080 signal from the full frame sensor.
I'm wondering how you determined that? I ask because in my own tests with the camera I certainly see aliasing (both in the raw still images as well as the 1080 video,) but I assumed that the aliasing in the video was caused or exaggerated by improper filtering in the scaling process (perhaps done intentionally to make the video look sharper). I'm not sure how I'd determine what actual algorithms were used in scaling, especially since the video's been compressed and potentially processed with god knows what other noise-reduction algorithm (thus hiding one of the remaining clues as to whether pixels have been averaged or just dropped), etc.
I ask really out of curiosity more than anything else, since for most of the same reasons as you I don't see a lot of uses of the camera for my applications.
Looking forward to seeing a GH1, though.
Jesse,
My understanding is that pixel binning is a sampling method to scale an image down to lower resolutions. Any image captured with an over-resolution sensor has to be scaled to get to standard HD resolutions. This itself isn't a problem, and many believe (self included) that it helps cut down sensor noise.
I think the issue you're concerned about is actually the alternate line sampling the 5DmkII video mode uses. Scanning alternating lines can introduce a moire pattern and false colors:
http://www.crunchgear.com/2008/12/02/for-image-quality-buffs-dslr-video-is-off-the-table/
Maybe only a problem if you're trying to film fine concentric circles, but I have no hands on experience with the 5DmkII.
Hi Clinton,
I know what binning is, and I've seen these images as well as my own that I've shot. Clearly there isn't sufficient filtering happening on the images of the 5DMk2 prior to being scaled to 1080P. But I see people stating that the camera is sampling alternate lines in video mode as if this were an established fact, but to state this based on a shot of a zone plate seems pretty spurious to me.
There are plenty of methods of scaling that could be happening inside this camera, several of which seem quite a bit more plausible from an engineering point of view (unless I'm missing something, which is why I'm asking the question) than what has been suggested.
Here be more footage for general consumption/picking apart. Rolling shutter makes an appearance, though it seems to be more of a cameo than in footage from the Canon or Nikon.
http://exposureroom.com/members/DVC.aspx/assets/38c5fbace36a4263b4d4fc21fafc341c/
good clip and skew is less than HV series. But the note say 1/60th shutter is slowest in 24p? Let's hope that's no true or a prototype limitation as that would be a dealkiller.
It also says "no pulldown was used by the camera," which we know is not possible. I'll believe that 1/60 shutter limitation when I see it with my own eyes.
Rolling shutter is definitely a problem, even with the new GH1, but they all exhibit the artifacting in different ways and to different degrees.
The 5DMkII clip is shot in exactly the wrong way.
I have a 5DMkII and use it extensively, but I also use it carefully and never truly 'handheld'. So much of the trouble is because of the form factor and how badly the skew shows when holding it like a DSLR is normally held for stills, and shooting shots with even a modest amount of telephoto will show ugly jello because of the unsteadyness of holding it with your hands.
Look at a few of the 5D2 clips on my vimeo page. One with my kids is even 24p converted with Optical Flow in Compressor. As a bonus, you can download the original QT's which are MUCH cleaner and higher res.
Oops, I apologize: here's the vimeo link
yeah, the more I think about it, a hard limit on shutter at 1/60th in 24p mode would probably require more CPU and DSP complexity than 1/48th.
http://vimeo.com/3269256
I'm guessing that even a small amount of skew will cause 3D camera tracking software to fail. This is very disappointing news for visual effects artists.
Hopefully, some software company will find a way to de-skew footage from CMOS sensors without introducing artifacts or significant softening.
I try very hard to avoid unnecessary resampling of the source footage to preserve the best possible quality, and I'm sure that if we could de-skew in post, you would have to do a slight blowup to get rid of black edges.
"The 5DMkII clip is shot in exactly the wrong way."
Barlow and Lannie, you're absolutely right. That's exactly what I think. Most of the videos demonstrating the jello effect are shot handhelp without some sort of stabilizer.
Sure the 5DMkII is not perfect but for the price it is phenomenal me thinks ;)
In my opinion, if you know your camera you should be able to get superb shots.
Look at this video to see how much you can get out of your 5DMkII.
http://www.vimeo.com/3162190
I do a lot of tracking/matchmoving with my HV20 and as long as you are very careful about movement you can get good tracks. If the jello effect is less with the GH1 then you're even better off.
Hey Stu -
Someone posted some 1080p footage from the GH1 that more clearly demonstrates the rolling shutter. I went CSI myself and stabilized it in AE. The effect is clearly more noticeable, but still not too bad.
http://vimeo.com/3577044
Source footage:
http://exposureroom.com/38c5fbace36a4263b4d4fc21fafc341c/
Stu:
Not sure if this interests you, but just as a proof of concept I spend an hour tonight making a quick program for my mobile phone that will display a set of fixed or custom grayscale screens. It enables me to consistently trick the 5D into using a specific exposure (I'm using manual Zeiss/Contax lenses so the aperture is already fixed) ... it's a lot easier than having to find an existing light source, anyway. Basically, I can quickly and dependably get 1/60s at 100 ISO in movie mode, or 1/50 in any other ISO (1/60 is only available at iso 100 ... don't have an F'in clue why).
http://cinema5d.com/viewtopic.php?f=14&t=1016&p=9288#p9288
I'll likely make a nicer PocketPC and iPhone version once I can test it out a bit more. What annoys me is that this should take about 1 day for Canon to fix (I do a fair amount of hardware development myself so I know how trivial it probably is), rather than me having to spend 20 seconds before every shot fiddling with my phone in front of a lens. I've decided that 'No firmware update for at least manual exposure', 'No more Canon purchases' ... tired of being treated like that after buying about $20K in lenses and bodies (and I already own an XHA1 if they're worried about losing a sale on that).
Stu,
Many thanks for the video samples.
"I do a lot of tracking/matchmoving with my HV20 and as long as you are very careful about movement you can get good tracks. If the jello effect is less with the GH1 then you're even better off."
Thanks Wes -- I did not know this was possible with an HV20.