Extent of rolling shutter reduction on 5D Mark III

[Andrew Reid]

[html]

[img]http://www.eoshd.com/wp-content/uploads/2012/03/rollingshutterdistortionofeos5dmark2.jpg[/img]

[url="http://mono-logue.air-nifty.com/monolog/2012/03/post-6b3f.html"]This Japanese test[/url] is quite revealing. Rolling shutter is reduced by half.

[url="http://www.eoshd.com/content/7384/extent-of-rolling-shutter-reduction-on-5d-mark-iii/"]Read full article[/url]

[/html]

[sandro]

pre production camera right... like the real deal is any better.

[heliorr]

Andrew did you tested the rolling shutter on the d800? because from what I'm seeing on the samples oh the markIII, I think the rolling shutter is better on the d800. from what I have seen on the Joy Ride, and from a little test I did with the camera on a presentation a week ago I think it really is better on the d800, there are almost no evidence of roling..

[Maxine]

Resolution looks even a little better on 5d II...

[sandro]

[quote author=Maxine link=topic=372.msg2316#msg2316 date=1331142072]
Resolution looks even a little better on 5d II...
[/quote]

I think that's the sharpening of the camera

[Andrew Reid]

There's more false detail on the 5D Mark II. No doubt 5D Mark III is better for cleanness of the resolution but disappointed it is not at GH2 level. On a $3500 DSLR, I mean c'mon FFS.

[MatthewP]

You sure it's not the camera's 720p mode?  ???

[rainman]

Hi Andrew.

Did you find any chance to review the effects of VAF-5D2 adapter on rolling shutter issues of 5D MII?

If it can reduce the rolling shutter issue on Mark II, it can be a good investment.

[Andrew Reid]

Why would an anti-aliasing filter help rolling shutter? No it doesn't have an affect.

[rainman]

So which way should we go?

5DMII+VAF 5D2 or 5DM3

[ted ramasola]

Looking at the stills of the ferris wheel, I saw the resolution of the mk3 similar to what a mk2 would look like with a VAF5d2.
Si I think its what canon did to reduce color aliasing and moire, they placed an agressive OLPF.

[Andrew Reid]

[quote author=rainman link=topic=372.msg2333#msg2333 date=1331154892]
So which way should we go?

5DMII+VAF 5D2 or 5DM3
[/quote]

Depends if you need the 5DM3 advantages and have a spare $3500 as opposed to $2500 (with VAF).

Advantages are:

- Better HDMI for monitoring, no 540p drop nonsense
- Sharp corners with wide lenses since no need to fix moire with VAF 5D2
- Better low light performance
- Timecode
- ALL-I codec and higher bitrates
- Reduced rolling shutter
- 3:2 screen
- Function button near shutter button
- Headphone jack
- Manual audio controls and meters (without need for Magic Lantern)
- Probably better DR

Those are off the top of my head, maybe some I've missed.

Worth $1000? I think so.

[RichST]

[quote author=ted ramasola link=topic=372.msg2336#msg2336 date=1331165338]
Looking at the stills of the ferris wheel, I saw the resolution of the mk3 similar to what a mk2 would look like with a VAF5d2.
Si I think its what canon did to reduce color aliasing and moire, they placed an agressive OLPF.
[/quote]

Surely not, an OLPF that strong would be unacceptable for stills. I suspect - but certainly can't prove - that they're doing something like 3x3 binning now instead of the old 3x1 binning + skip two lines. That would jive with the reduced aliasing and better low light performance we're seeing as well as the minor differences in resolution. The sensor also looks to be scanning faster which explains the reduced rolling shutter (they didn't double the number of output channels for no reason). All in all these look like simple, common sense improvements, nothing revolutionary but still good to have

[TJB]

If I didn't have a 5DMKII with VAF5D2 I would definitely go for the 5DMKIII for all the reasons Andrew mentioned plus some personal ones - including -
-I shoot stills and video and it's pain to remove the VAF5D2 to shoot stills.
-Focus on every lens I have behaves differently with the filter inserted. For example on the EF 24-105mm lens at 24mm - infinity focus is actually macro on the opposite end of the focus barrel.
-Even slight reframing with the zoom requires a new focus check.
I'm in no hurry to upgrade. But if I was building a new kit, I would go for the MKIII. If I wanted the magnificent 5X crop mode I would definitely go for the GH2.
Getting the EF 24-105mm kit lens is highly recommended. It's image stabilisation is fantastic - very quiet and easily usable on the long end. f4 on a full frame body has a similar DOF as the GH2 with an f2 lens. I can't chase focus on moving subjects with a larger aperture than f4 anyway.

[Andrew Reid]

[quote author=RichST link=topic=372.msg2340#msg2340 date=1331187156]
[quote author=ted ramasola link=topic=372.msg2336#msg2336 date=1331165338]
Looking at the stills of the ferris wheel, I saw the resolution of the mk3 similar to what a mk2 would look like with a VAF5d2.
Si I think its what canon did to reduce color aliasing and moire, they placed an agressive OLPF.
[/quote]

Surely not, an OLPF that strong would be unacceptable for stills. I suspect - but certainly can't prove - that they're doing something like 3x3 binning now instead of the old 3x1 binning + skip two lines. That would jive with the reduced aliasing and better low light performance we're seeing as well as the minor differences in resolution. The sensor also looks to be scanning faster which explains the reduced rolling shutter (they didn't double the number of output channels for no reason). All in all these look like simple, common sense improvements, nothing revolutionary but still good to have
[/quote]

Nice analysis. What trick would you say the GH2 has up her sleeve to get the video resolution it does from a 18MP native chip? What is the next step up from 3x3 binning? My guess is that the GH2's processor is reading out a 4MP or even 8MP 4K image from the sensor and downscaling intelligently to 2MP on the image processor itself. I cannot think of any other way it would be better than so many cameras that do binning on the sensor like the NEX 7, 5D Mark III, Nikon D800, etc.

[bg2b]

[quote author=Andrew Reid - EOSHD link=topic=372.msg2344#msg2344 date=1331206890]
Nice analysis. What trick would you say the GH2 has up her sleeve to get the video resolution it does from a 18MP native chip? What is the next step up from 3x3 binning? My guess is that the GH2's processor is reading out a 4MP or even 8MP 4K image from the sensor and downscaling intelligently to 2MP on the image processor itself. I cannot think of any other way it would be better than so many cameras that do binning on the sensor like the NEX 7, 5D Mark III, Nikon D800, etc.
[/quote]
I have no idea what the GH2 does, but perhaps you could get an idea by looking at how it renders a zone plate.  Different downscaling methods tend to produce different "signatures", ala [url=http://www.cambridgeincolour.com/tutorials/image-resize-for-web.htm]http://www.cambridgeincolour.com/tutorials/image-resize-for-web.htm[/url] and a few other million references.

[Tzedekh]

[quote author=Andrew Reid - EOSHD link=topic=372.msg2344#msg2344 date=1331206890]
What trick would you say the GH2 has up her sleeve to get the video resolution it does from a 18MP native chip? What is the next step up from 3x3 binning? My guess is that the GH2's processor is reading out a 4MP or even 8MP 4K image from the sensor and downscaling intelligently to 2MP on the image processor itself. I cannot think of any other way it would be better than so many cameras that do binning on the sensor like the NEX 7, 5D Mark III, Nikon D800, etc.[/quote]
If the GH2 could read out 8 MP -- in effect, QuadHD (3,840 x 2,160) -- and then downscale on the fly to 1080p, wouldn't it be able to skip the downscale and output QuadHD directly? The JVC HMQ10 can do 4K2K at 60p at a data rate of 144 Mbps, well within the rate the Driftwood hack proved possible. That 144 Mbps comprises four 36-Mbps streams, which is roughly the midpoint of Panasonic's AVC-Ultra LongG range (25-50 Mbps) for 10-bit 4:2:2. Multiply that by 4 to get QuadHD and you get 100-200 Mbps. So, if properly encoded, 144 Mbps should theoretically be enough to handle 3,840 x 1,920 at 10 bits and 4:2:2.

[RichST]

[quote author=Andrew Reid - EOSHD link=topic=372.msg2344#msg2344 date=1331206890]
[quote author=RichST link=topic=372.msg2340#msg2340 date=1331187156]
[quote author=ted ramasola link=topic=372.msg2336#msg2336 date=1331165338]
Looking at the stills of the ferris wheel, I saw the resolution of the mk3 similar to what a mk2 would look like with a VAF5d2.
Si I think its what canon did to reduce color aliasing and moire, they placed an agressive OLPF.
[/quote]

Surely not, an OLPF that strong would be unacceptable for stills. I suspect - but certainly can't prove - that they're doing something like 3x3 binning now instead of the old 3x1 binning + skip two lines. That would jive with the reduced aliasing and better low light performance we're seeing as well as the minor differences in resolution. The sensor also looks to be scanning faster which explains the reduced rolling shutter (they didn't double the number of output channels for no reason). All in all these look like simple, common sense improvements, nothing revolutionary but still good to have
[/quote]

Nice analysis. What trick would you say the GH2 has up her sleeve to get the video resolution it does from a 18MP native chip? What is the next step up from 3x3 binning? My guess is that the GH2's processor is reading out a 4MP or even 8MP 4K image from the sensor and downscaling intelligently to 2MP on the image processor itself. I cannot think of any other way it would be better than so many cameras that do binning on the sensor like the NEX 7, 5D Mark III, Nikon D800, etc.
[/quote]

I'm at a loss on that one. I don't think they're getting 4K worth of data and downscaling, resolution should be higher if that were the case. FWIW I don't think the Nikon and Sonys are using full binning of all pixels on the sensor, if that were so then those shimmering fine lines that dance and change color shouldn't be there. But we know from Sony's presentations that they [i]do[/i] use 2x2 binning. My guess is that they are either binning groups then skipping over some columns or they're just binning one color at a time to get a reduced Bayer pattern. Who knows, Panasonic could be using a similar technique but are maybe just adding a little extra secret sauce to the formula, like sampling more greens (that's just a[i] total[/i] guess and probably not correct). Whatever it is it's not only better than Sony Canon et al it's also better than their own non GH-series sensors.

The downsampling methods shown on that Cambridge site all assume you are sampling every pixel on the sensor and reducing it from there, that would require a full sensor scan in video mode which no consumer camera can currently do (the Nikon One is fast enough but they don't make use of it). I take that back: JVC's ugly little camcorder can evidently do that with a tiny chip but the processors aren't fast enough to properly deal with all that data, it has to get butchered in the process of converting it to video

[bg2b]

[quote author=RichST link=topic=372.msg2354#msg2354 date=1331238627]
The downsampling methods shown on that Cambridge site all assume you are sampling every pixel on the sensor and reducing it from there, that would require a full sensor scan in video mode which no consumer camera can currently do (the Nikon One is fast enough but they don't make use of it). I take that back: JVC's ugly little camcorder can evidently do that with a tiny chip but the processors aren't fast enough to properly deal with all that data, it has to get butchered in the process of converting it to video
[/quote]
Doesn't the GH2 have some sort of parallel on-sensor A/D?  Perhaps they're doing some amount of digital combining on-sensor before actually reading off the data.  (I have no idea what they're doing of course, but it seems plausible that having a bunch of highly parallel logic on the sensor might enable them to reduce the data that they have to process off-sensor by a large amount.  It would be interesting to turn someone like Chipworks loose on one of the sensors.)