Whether the 3×3 binning on 5D Mark III will cause resolution loss？

[gelaxstudio]

AS the title shows~
For my opinion，the 3×3 binning is much like a digital low pass Filter .By binning pixel ，it cloud reduce Moire，but like the low pass Filter ，it will cause resolution loss in return.
When I saw the ISO test video shooted by 5DM3,it showed resolution loss ,but I did't think it's caused by NR or out of focus.After all,the video was came from Canon Japan,which mean it is highly credible.
What's more ,as it's cuased by binning pixel,the high bit rate mode like ALL-I won't be able to make up the lost resolution,and sad to find out this function can't be disable if user wanted.

I am not a expert in this aspect,so I can't sure my opinion is absolutely correct.
So let's figure it out.

 

[HurtinMinorKey]

If you have a 1080 clip form the mkiii could you please post it?

 

[Axilrod]

AS the title shows~
For my opinion，the 3×3 binning is much like a digital low pass Filter .By binning pixel ，it cloud reduce Moire，but like the low pass Filter ，it will cause resolution loss in return.
When I saw the ISO test video shooted by 5DM3,it showed resolution loss ,but I did't think it's caused by NR or out of focus.After all,the video was came from Canon Japan,which mean it is highly credible.
What's more ,as it's cuased by binning pixel,the high bit rate mode like ALL-I won't be able to make up the lost resolution,and sad to find out this function can't be disable if user wanted.

I am not a expert in this aspect,so I can't sure my opinion is absolutely correct.
So let's discuss and find the answer .

I don't think the pixel binning is an issue with the 5DIII, the 22.3MP works out to be 1920 x 1080 when divided. And I wouldn't trust what you saw on the web. No matter how good an encoder the website says they have, there is always some loss when uploading to the web.

 

[HurtinMinorKey]

I found a link to some higher bit-rate mkiii footage over at cinema 5D. Once you click on the link, you can download it with the tab on the right.

https://skydrive.live.com/?cid=a218911273df094b&id=A218911273DF094B!365&sff=1

 

[mrmarks]

I don't think the pixel binning is an issue with the 5DIII, the 22.3MP works out to be 1920 x 1080 when divided.

Hi Axilrod, could you explain what you meant by this? Thanks

 

[gelaxstudio]

I don't think the pixel binning is an issue with the 5DIII, the 22.3MP works out to be 1920 x 1080 when divided. And I wouldn't trust what you saw on the web. No matter how good an encoder the website says they have, there is always some loss when uploading to the web.

https://vimeo.com/37879608
This is the link to the ISO teat video ,and it's a comparison between 5D2 and 5D3,you should check it out.
The resolution of 5D3 is visible lower than 5D2,they should be the same NR settings and it's an official video edited by Canon Japan!
So I can't figure out other explanation for this issue.

 

[DavidRiesenberg]

That video is a very poor example of comparison. Most notably is what appears to be different lenses used. No information about F-stop, camera profiles and what appears to be post processed footage. The 5DII looks artificially sharpened while the 5DIII looks like it has a NR Filter applied to it.

 

[gelaxstudio]

That video is a very poor example of comparison. Most notably is what appears to be different lenses used. No information about F-stop, camera profiles and what appears to be post processed footage. The 5DII looks artificially sharpened while the 5DIII looks like it has a NR Filter applied to it.

Why Canon did that？ Making a joke ?
Why no try to make the 5D3 perform better than 5D2 in all aspects?
The 5D3 is supposed to be better than 5D2 in resolution,because it has upgraded to 22 million pixels~

 

[LetTheRightLensIn]

That video is a very poor example of comparison. Most notably is what appears to be different lenses used. No information about F-stop, camera profiles and what appears to be post processed footage. The 5DII looks artificially sharpened while the 5DIII looks like it has a NR Filter applied to it.

Why Canon did that？ Making a joke ?
Why no try to make the 5D3 perform better than 5D2 in all aspects?
The 5D3 is supposed to be better than 5D2 in resolution,because it has upgraded to 22 million pixels~

They appear to be so obsessed with hiding noise they'd rather put up noisless but waxy,soupy looking awful footage (and stills). I don't get it. I just hope, that as in the past, once the cams get into real users hands everything looks as it should once again.

 

[Jeff Gibbs]

The thing that concerns me about that comparison is that yes the 5d Mark III looks WORSE overall than the Mark II, despite the improvement in artifacts and latitude, but that whatever the potential difference in lens used, etc. a new generation camera ought to be a stunner, a clear winner, not at all debatable. If this is the final product I am reconsidering unless there is some drastic explanation.

 

[Stephen Melvin]

Do we have confirmation that the Mk III is binning pixels? Everything Canon has said implies that moiré is being removed by the DIGIC 5+ processor, and mentions nothing about binning.

Also, if the sensor is being binned, why is the maximum sensitivity of the movie mode lower than than that of stills mode? Binning should allow for higher ISO sensitivity, not lower.

 

[Policar]

Do we have confirmation that the Mk III is binning pixels? Everything Canon has said implies that moiré is being removed by the DIGIC 5+ processor, and mentions nothing about binning.

Also, if the sensor is being binned, why is the maximum sensitivity of the movie mode lower than than that of stills mode? Binning should allow for higher ISO sensitivity, not lower.

The idea that it uses 3x3 pixel binning is based on some quote from a Canon spokesperson regarding the video in the 1DX, I believe. I think he said something along the lines of "all pixels are being read out on the 1DX" and then someone said the 5DIII did something similar--and now for whatever reason everyone assumes it's 3x3 binning.

That said, canceling out aliasing in post is like virtually impossible without absolutely destroying resolution. And the 5DII only skipped pixels in one axis it seems, probably binning them in the other. So it's reasonable to conclude (or at least hope) its not 3x3 pixel skipping, which should be much worse....

This reasoning is still just a bit crazy. Not just because the idea is based on quote regarding a different camera without a neat 1080p divisor res'ed sensor but moreso because the 5DIII's sensor doesn't divide into 1920X1080 neatly as the c300's does. Bayer is 1:2:1 in terms of how R:G:B are represented. Apply that to 5760X3840 and you have 1440X810 for two channels and 2880X1620 for the other. Not 1080p at all. If the c300 is doing what people label 2x2 pixel binning, it's getting true 1080p out of it. If the 5DIII is doing the same process but 3x3, it's getting upscaled "810p" out of it.

 

[LetTheRightLensIn]

Do we have confirmation that the Mk III is binning pixels? Everything Canon has said implies that moiré is being removed by the DIGIC 5+ processor, and mentions nothing about binning.

Also, if the sensor is being binned, why is the maximum sensitivity of the movie mode lower than than that of stills mode? Binning should allow for higher ISO sensitivity, not lower.

The idea that it uses 3x3 pixel binning is based on some quote from a Canon spokesperson regarding the video in the 1DX, I believe. I think he said something along the lines of "all pixels are being read out on the 1DX" and then someone said the 5DIII did something similar--and now for whatever reason everyone assumes it's 3x3 binning.

That said, canceling out aliasing in post is like virtually impossible without absolutely destroying resolution. And the 5DII only skipped pixels in one axis it seems, probably binning them in the other. So it's reasonable to conclude (or at least hope) its not 3x3 pixel skipping, which should be much worse....

This reasoning is still just a bit crazy. Not just because the idea is based on quote regarding a different camera without a neat 1080p divisor res'ed sensor but moreso because the 5DIII's sensor doesn't divide into 1920X1080 neatly as the c300's does. Bayer is 1:2:1 in terms of how R:G:B are represented. Apply that to 5760X3840 and you have 1440X810 for two channels and 2880X1620 for the other. Not 1080p at all. If the c300 is doing what people label 2x2 pixel binning, it's getting true 1080p out of it. If the 5DIII is doing the same process but 3x3, it's getting upscaled "810p" out of it.

Are you sure 3x3 means 810p??

Wouldn't each block of 3x3 goes to one pixel so you get like 5 greens and 2 reds and 2 blues of info going into a pixel of the 1920x1080 output?

GBG
RGR
GBG to one true color pixel perhaps?


Some place Canon said something about, on sensor, it combines it down from the entire frame and send 1920x1080 to the digic, I thought.

 

[peederj]

Digital signal processing is too deep a topic for crowdsourced wisdom. You will have to wait for real experts to analyze the results and shed light on what's going on. If they are willing to.

 

[Stephen Melvin]

Over on DPR, Joseph Wisniewski is talking about the "3.5x3.5" binning that the D800 does. It didn't occur to me that you could bin by half.

http://forums.dpreview.com/forums/read.asp?forum=1021&message=40975606