R5 AA/OLPF filter type?

The R5 contains an anti-aliasing optical low pass filter, but I have not been able to find details on this. I think it was widely speculated that the R5 got the 16-point OLPF from the EOS-1D X Mark III, but I don't think that is the case. I think the R5 actually has an asymmetric OLPF. But I can't find more info on that.

Does anyone have more info on the OLPF in the R5, and the R3, R6, R7 as well?

Ephemeris wrote:

Other than remembering some talk existed around launch I dont know any more I'm sorry.

I don't know if posting in Photographic Science and Technology group would be helpful in parallel.

Thanks. I have a thread going there, and was looking for more info:

https://www.dpreview.com/forums/post/66820426

BTW, that talk included Canon claiming that the R5 had higher resolution than the 5DsR, but no it doesn't. The R7 does match it on average though.

Ephemeris wrote:

Sorry I had missed your post in that group, although I've now read it through.

I assume the surface structure of the sensor doesn't lead any clues? To rule out any physical structure.

My current thinking is that the asymmetric AA filter is not a bug but a feature. The R6ii seems to have a conventional symmetrical 4-point AA filter, and it has DPAF, so it's not baked in to the DPAF sensor structure (at 6um pixel scale anyway).

I think Canon needs good anti-aliasing of vertical lines to prevent DPAF errors--otherwise the camera could misfocus on aliased false details. So maybe this asymmetrical filter is the new improved AA filter technology on the R5 that trickled down to the R7. But it seems there is still some weak anti-aliasing on horizontal edges because there is no dead cat bounce above 1 cy/px on the R7 vs the 5DsR (I didn't post those extended SFR domain graphs yet).

I don't have other R-series bodies to cross test, so I don't know if the R3/R6 are similar to the 6Dii or R5/R7. I'd also be interested in any info on their AA/OLPF filter type.

Ephemeris wrote:

Bigger wrote:

Ephemeris wrote:

Sorry I had missed your post in that group, although I've now read it through.

I assume the surface structure of the sensor doesn't lead any clues? To rule out any physical structure.

My current thinking is that the asymmetric AA filter is not a bug but a feature. The R6ii 6Dii seems to have a conventional symmetrical 4-point AA filter, and it has DPAF, so it's not baked in to the DPAF sensor structure (at 6um pixel scale anyway).

I think Canon needs good anti-aliasing of vertical lines to prevent DPAF errors--otherwise the camera could misfocus on aliased false details. So maybe this asymmetrical filter is the new improved AA filter technology on the R5 that trickled down to the R7. But it seems there is still some weak anti-aliasing on horizontal edges because there is no dead cat bounce above 1 cy/px on the R7 vs the 5DsR (I didn't post those extended SFR domain graphs yet).

I don't have other R-series bodies to cross test, so I don't know if the R3/R6 are similar to the 6Dii or R5/R7. I'd also be interested in any info on their AA/OLPF filter type.

Thanks bigger. I think the data I've seen posted by you and a few other members is interesting and shows good comparison. They (Canon) made quite a big deal about the filter(s) before launch but (I don't recall exactly) but it seems like what you have found is different.

That alone is a little odd don't you think?

Yes, I think it's odd that I can't find any details on the new filter on the R5 sensor, even after 3 years. And I have not seen any other independent tests, other than the optydene.pl review of the R7. So if you have links, please share.

I did some more digging in the DPR studio scenes, and ran the raw files through MTF Mapper. I found that lots of late model Canon sensors have the asymmetric filter like the R5 & R7 as I found, including the R6, R6ii and R10, but not the R & RP which are symmetric. The R3 is just slightly asymmetric as is the 1Diii so it may have the same 16-point filter. All of the M's are asymmetric as are the T6's and T7's. Interestingly, the 70D and 80D are symmetric, but the 77D and 90D are asymmetric. The 1Dii, 5Ds and 5Div are symmetric, and the 5DsR has no AA filtering.

I'll post more details to the other thread.

P.S. I corrected an error in my original post above. I meant to say the 6Dii has a symmetrical filter, not the R6ii.

Atta-Boy wrote:

Bigger wrote:

I think Canon needs good anti-aliasing of vertical lines to prevent DPAF errors--otherwise the camera could misfocus on aliased false details.

if I may ask a question - why 'd DPAF errors occur when multitude of DPAF sensels participate ( are behind each of the focusing points ) and why 'd CDAF ( I think - and I of course might be very much wrong - it is some equivalent aggregation of a number of sensels participating in the focusing operation ) in cameras w/o AA filters not misfocus on aliased false details ?

Because CDAF is dumb, and DPAF is smart, and sometimes too smart by half. CDAF can misfocus when it finds a local maxima in contrast and stops looking, so my 5DsR is prone to this too, but it's rare. Symmetrical AA filters give more defense against this.

DPAF only focuses on vertical edges, so it only needs AA filtering applied in the horizontal direction. I suspect that Canon found that DPAF benefits from relatively strong AA filtering in the horizontal direction, so they have been applying it to all new bodies, but most seem to have relatively weak AA filtering in the vertical direction. This is exactly opposite of the orientation of the asymmetric AA filter by Nikon, so I assume it is related to DPAF.

Atta-Boy wrote:

Bigger wrote:

I think it was widely speculated that the R5 got the 16-point OLPF from the EOS-1D X Mark III,

I subjectively think that if it were the case then Canon marketing will be officially writing somewhere - look what we got here, but they did not...

Yes, I don't think this is the case now. The R5 seems to have the strong asymmetric AA filter similar to the other late model Canon DPAF sensors. Maybe it has an incremental improvement, but the R5 SFR plot does not look like the 1Diii with the 16-point filter. The R3 SFR does look similar, though.

Atta-Boy wrote:

Bigger wrote:

Atta-Boy wrote:

Bigger wrote:

I think Canon needs good anti-aliasing of vertical lines to prevent DPAF errors--otherwise the camera could misfocus on aliased false details.

if I may ask a question - why 'd DPAF errors occur when multitude of DPAF sensels participate ( are behind each of the focusing points ) and why 'd CDAF ( I think - and I of course might be very much wrong - it is some equivalent aggregation of a number of sensels participating in the focusing operation ) in cameras w/o AA filters not misfocus on aliased false details ?

Because CDAF is dumb, and DPAF is smart, and sometimes too smart by half. CDAF can misfocus when it finds a local maxima in contrast and stops looking

I am sorry but so does PDAF ...
and it still does not answer the question how so many DPAF sensels ( because clearly it is not just one sensel /a pair behind it/) in a situation when clearly data is aggregated from a lot of them can be wrong at once

Canon DPAF doesn't have sensels, it uses the dual pixels on the sensor.