EOS 6D Anti-Aliasing Filter

[compupix]

With the advent of Mosaic Engineering's VAF-6D Optical Anti-Aliasing Filter, one would hope that Canon would come up with a firmware solution... if possible. Maybe the 6D doesn't have the processing power to do it.
But, when a third-party offers a fix for a problem... you've really got a problem.
http://www.mosaicengineering.com/products/vaf/6d.html
Here's the demo video:
https://vimeo.com/58335679

Thanks,
Drew

 

[compupix]

I watched the video. It looked impressive comparing with and without the filter.
The question is, why is the 5D III so superior to the 6D in this area? Is it processing power or something else?

Thanks,
Drew

 

[mdm]

an anti-aliasing filter I believe is little more than a film which has a slightly matte finish to it.. it's purpose is to blur visual information so that moirè is less likely to happen

in audio I think this is called dithering.. noise is added to a signal so that symmetrically repeated errors don't occur while digitizing

 

[Marsu42]

The question is, why is the 5D III so superior to the 6D in this area? Is it processing power or something else?

I recently read (on CR, but I cannot find the link anymore) that the 5d3 has an entirely different sampling algorithm to downsize the sensor resolution to video, hence the big difference in video moiré.

That and the stronger 5d3 aa filter (vs. 5d2) would also explain that some reviews state that the 6d video is sharper than the 5d3, though at the high cost of more aliasing that has to be removed in post.

 

[emag]

But, when a third-party offers a fix for a problem... you've really got a problem.

I understand your point of view, but I don't consider it a 'problem', I believe the camera was intentionally designed that way for still photos, it was not intended to be optimized for video. What you consider a third-party fix for a problem I consider a third-party enhancement.....an aftermarket accessory. If it does the job it's a good thing for all.

 

[titokane]

The 5D3 is, when cropped from 3:2 to 16:9, is essentially a QFHD sensor -- exactly 4 times the size of standard 1080p HD. The reason it looks good in 1080p mode is because the camera only has to combine 4 pixels into a single pixel (pixel binning). This is the exact same thing the C300 does to look so good. The 1DX has a smaller sensor that isn't a simple multiple, but it has very powerful processors to downsample in a high quality way. The 6D has a smaller sensor but doesn't have the processing power to downsample correctly (presumably) so Canon just throws out various lines to create the smaller image. It's less processor intensive but results in interference patterns. That's why the 5D3 looks great with moire, the 1DX looks almost as good, and the 6D still has the same problems that the 5D2 had.

If it turns out that the processor is strong enough to downsample properly, it could possibly be adjusted in a firmware update that rewrites the downsampling algorithm. Is it likely that Canon will do that? I'm not counting on it.

 

[compupix]

Thanks Titokane!

 

[Badger]

But, when a third-party offers a fix for a problem... you've really got a problem.

I understand your point of view, but I don't consider it a 'problem', I believe the camera was intentionally designed that way for still photos, it was not intended to be optimized for video. What you consider a third-party fix for a problem I consider a third-party enhancement.....an aftermarket accessory. If it does the job it's a good thing for all.

Emag,
I really don't want to give Canon a pass on this. It IS a problem. I video tapped my daughter's band recital the other day. There were folded up bleachers in the background. The Morié was so bad it was embarrassing. Don't get me wrong, I love the camera but when I got it, my intention was to do still and video. Yes I did consider the 5D MkIII, and no, I didn't have the money for it.

 

[rs]

The 5D3 is, when cropped from 3:2 to 16:9, is essentially a QFHD sensor -- exactly 4 times the size of standard 1080p HD. The reason it looks good in 1080p mode is because the camera only has to combine 4 pixels into a single pixel (pixel binning). This is the exact same thing the C300 does to look so good. The 1DX has a smaller sensor that isn't a simple multiple, but it has very powerful processors to downsample in a high quality way. The 6D has a smaller sensor but doesn't have the processing power to downsample correctly (presumably) so Canon just throws out various lines to create the smaller image. It's less processor intensive but results in interference patterns. That's why the 5D3 looks great with moire, the 1DX looks almost as good, and the 6D still has the same problems that the 5D2 had.

If it turns out that the processor is strong enough to downsample properly, it could possibly be adjusted in a firmware update that rewrites the downsampling algorithm. Is it likely that Canon will do that? I'm not counting on it.

Spot on, except the 22mp 5D3 has 3 times the horizontal resolution and 3 times the vertical resolution of 1080p, so it combines 9 sensor pixels to make every one video pixel.

 

[titokane]

The 5D3 is, when cropped from 3:2 to 16:9, is essentially a QFHD sensor -- exactly 4 times the size of standard 1080p HD. The reason it looks good in 1080p mode is because the camera only has to combine 4 pixels into a single pixel (pixel binning). This is the exact same thing the C300 does to look so good. The 1DX has a smaller sensor that isn't a simple multiple, but it has very powerful processors to downsample in a high quality way. The 6D has a smaller sensor but doesn't have the processing power to downsample correctly (presumably) so Canon just throws out various lines to create the smaller image. It's less processor intensive but results in interference patterns. That's why the 5D3 looks great with moire, the 1DX looks almost as good, and the 6D still has the same problems that the 5D2 had.

If it turns out that the processor is strong enough to downsample properly, it could possibly be adjusted in a firmware update that rewrites the downsampling algorithm. Is it likely that Canon will do that? I'm not counting on it.

Spot on, except the 22mp 5D3 has 3 times the horizontal resolution and 3 times the vertical resolution of 1080p, so it combines 9 sensor pixels to make every one video pixel.

Yes! How embarrassing! I've been saying it's 2x2 binning (like the C300 does) instead of 3x3 binning in a few posts. Thanks for the correction! But if anybody's wondering, the logic still stands. Evenly combining those 9 pixels instead of skipping various lines is a much better way to downsample, and leads to much less moire.

 

[dmosier]

I use DSLRs almost exclusively for video so I really want the 6D to be my camera of choice but the moire really bothers me. I haven't used it personally so, unfortunately, the only way for me to analyze its performance is through videos posted online. This is not ideal as I often have no idea what camera settings were used, and there is always the possibility that the image is further affected when uploaded online.

I have used the MkII, 7D, and 60D (MkIII a little but not enough to say I am familiar with it), and I find the moire/aliasing in those cameras to be perfectly acceptable for my needs. But, based solely on what I've seen online, the 6D looks worse than those cameras. All I can find is people stating how much worse the 6D is than the MkIII, which I should expect for $1000 difference. But I still can't find anyone with an informed, personal, experience with the 6D compared to those other DSLRs that can tell me whether or not the moire is the same as, better, or worse.

 

[aroo]

If the lines skipped were all at the top and bottom of the frame, the moire could be less, but with an effectively higher crop factor on the image. Might change overall IQ significantly?

 

[iP337]

Are saying for sure the 5D3 doesn't line skip, it pixel binds like the GH2/GH3 and C300?

Panasonic has actually admitted to using 2x2 pixel binding plus some further pixel combining to get their 1080p resolution, which is not only low in morie but also still sharp:

DE: So you feel that the GH2's video quality is better than the 5D Mark II's?

Panasonic: Yes, it's not our engineers saying so. Many professionals have said that the GH2's video was better, even with a smaller sensor.

But compared to a professional camcorder, Micro Four Thirds is much better. It depends on the signal processing, but we also have so many pixels that we can do various kinds of pixel combining. Maybe this is some of why we're better than the 5D Mark II: we do better pixel combining. This is maybe why Canon prefers just an 8-megapixel sensor on their C300, because they have to combine just 2 x 2 pixels. Also, it's possible to read all pixels in real time.

DE: You mentioned that with the C300, they read out the entire frame in real time. So in a large format camera like the GH3, you are reading only some of the pixels; you don't actually read the whole array…

Panasonic: Yes, it's not possible to read out the whole array at 60p. So this means we have to do some pixel combining. We combine 2 x 2 pixels on the sensor, so that makes just one quarter the data that we have to read out. This makes it possible to read out at 60p. Then we combine again to get the 2 megapixel video image.

...and it would seem Nikon's new Toshiba sensor does the same too ( http://www.eoshd.com/content/9586/is-the-cheap-nikon-d5200-a-better-option-than-d800-for-video-no-moire-aliasing-and-good-detail ) getting the same low morie and sharp image.

If anyone has a link confirming the 5D3 uses pixel binding please share because I am more inclined to believe that Canon just slapped on a stronger AA filter because the results look a lot like the 5D2 and 6D with these adapters which is lower morie and sharpness.

 

[dmosier]

I would like to add something I read in another thread about the 6D and this filter that is worth pointing out here.

That video is presented by a company that is selling a filter designed to "fix" moire and aliasing. So who is to say they didn't make the "before" video look as poor as possible in order to make their filter seem like an indispensable item? The 6D may very well look that lousy by default. How can I tell? I've never used one. But there is also no way to tell that they didn't tweak the camera settings to make it look like garbage just to sell their filter.

So that video alone is not sufficient to judge how good or bad the 6D really is.

 

[compupix]

Here's another video comparing the EOS 5D III, II, and 6D:
http://canoneos5dmarkiii.org/canoneos/canon-full-frame-shootout-canon-eos-6d-vs-canon-eos-5d-mark-iii-vs-canon-eos-5d-mark-ii
This video seems to match the video's performance from the company that is selling the filter.

 

[iP337]

I would like to add something I read in another thread about the 6D and this filter that is worth pointing out here.

That video is presented by a company that is selling a filter designed to "fix" moire and aliasing. So who is to say they didn't make the "before" video look as poor as possible in order to make their filter seem like an indispensable item? The 6D may very well look that lousy by default. How can I tell? I've never used one. But there is also no way to tell that they didn't tweak the camera settings to make it look like garbage just to sell their filter.

So that video alone is not sufficient to judge how good or bad the 6D really is.

Yeah that was me lol.

I'm not saying the 5D3 isn't better than a 6D and it seems practical to pixel bind with the native resolution of the 5D3 but I am seeing very similar if not the same results with the 6D and this third party AA filter that I see with native 5D3 video footage. I just want to know for sure if 5D3 really pixel binds, everyone keeps saying it does but I've yet to see anything more convincing than here-say and rumors.

 

[Marsu42]

I'm not saying the 5D3 isn't better than a 6D and it seems practical to pixel bind with the native resolution of the 5D3 but I am seeing very similar if not the same results with the 6D and this third party AA filter that I see with native 5D3 video footage.

Terrific - just one detail though: "VAF-6D Moire / Aliasing Filter - Your Price: $365.00" ... and thinking of the current 5d3-6d price difference, the tons of things that make the 5d3 a more capable camera ... well, you know :-o ...

 

[iP337]

I'm not saying the 5D3 isn't better than a 6D and it seems practical to pixel bind with the native resolution of the 5D3 but I am seeing very similar if not the same results with the 6D and this third party AA filter that I see with native 5D3 video footage.

Terrific - just one detail though: "VAF-6D Moire / Aliasing Filter - Your Price: $365.00" ... and thinking of the current 5d3-6d price difference, the tons of things that make the 5d3 a more capable camera ... well, you know :-o ...

Very true, that's why I keep saying the 6D isn't worth it until it hits $1500 or less.

 

[studio1972]

The question is, why is the 5D III so superior to the 6D in this area? Is it processing power or something else?

Thanks,
Drew

The 5D3 sensor is precisely 3 x the horizontal resolution of 1080p video, so all they have to do is combine each nine pixels into 1.

 

[iP337]

The question is, why is the 5D III so superior to the 6D in this area? Is it processing power or something else?

Thanks,
Drew

The 5D3 sensor is precisely 3 x the horizontal resolution of 1080p video, so all they have to do is combine each nine pixels into 1.

Right, but are they actually doing that? Everyone keeps pointing out how easy and practical it is to do but I've never heard or seen anything else supporting the fact they are doing that. I want to believe it, I really really hope they are but I still don't have proof.