The EOS 80D Replacement to be a Big Leap Forward [CR2]

[fullstop]

@Neuro
thanks for the explanation. need to mull it over some more to fully understand all ramifications.

especially why i notice the phenomenon of "too small af marker boxes" at times on EOS M (1st gen), but not really on the 5D3. some combination of aps- c vs FF sensor, slower EF-M lenses vs 5D3 on which i generally use f/2.8 lenses or faster) and/or differences in scenes i capture with the 2 systems.

will also be interesting to find out whether EOS M50 with DP-AF and more AF points brings an improvement in practical terms in this regard.

i picked one up just yesterday - funnily and totally co-incidentally exactly at the time when it was questioned here, whether i still qualify as *current* Canon customer at all or just as "dry dock Cano-whiner". M50 is actually for my daughter and she too has noticed the af boxes on EOS M (with 22/2.0) not always matching af-sensitive area. if she does not complain about the issue any more, i consider the issue solved ... no matter how it may have been accomplished in the face of "unforgiving laws of optics".

 

[3kramd5]

That's the first I've seen a picture of a 1DXII PDAF sensor, I think.

That’s the 1Dx / 5D3 PDAF assembly.
The 1Dx II presumably has a different configuration of sensors to facilitate f/8 sensitivity at each location.

i picked one up just yesterday - funnily and totally co-incidentally exactly at the time when it was questioned here, whether i still qualify as *current* Canon customer at all or just as "dry dock Cano-whiner".

I stand corrected. You indeed are a current customer. Note, nobody said “dry dock Cano-whiner.”

daughter and she too has noticed the af boxes on EOS M (with 22/2.0) not always matching af-sensitive area.

Having never used DPAF, I don’t know how it works relative to orientation. I presume they construct lines analogous to dedicated sensors, not actually use a single photo site, but that’s a total guess. If so, what are the orientations of the lines, and does that affect their sensitivity?

If they are constrained directionally, that could explain the presence of selectable AF point which finds no subject.

With other on sensor schemes such as sony’s, all the PDAF lines are horizontal, I believe.

 

[neuroanatomist]

Having never used DPAF, I don’t know how it works relative to orientation. I presume they construct lines analogous to dedicated sensors, not actually use a single photo site, but that’s a total guess. If so, what are the orientations of the lines, and does that affect their sensitivity?

If they are constrained directionally, that could explain the presence of selectable AF point which finds no subject.

The DPAF pixels are indeed constrained directionally, the hemi-pixel divisions are all in the same orientation. I really think Canon needs to explore the idea of QPAF - quad-pixels (while a technical challenge, I think that would be than alternating/mixing the orientation of the pixel divisions).

 

[Durf]

Having never used DPAF, I don’t know how it works relative to orientation. I presume they construct lines analogous to dedicated sensors, not actually use a single photo site, but that’s a total guess. If so, what are the orientations of the lines, and does that affect their sensitivity?

If they are constrained directionally, that could explain the presence of selectable AF point which finds no subject.

The DPAF pixels are indeed constrained directionally, the hemi-pixel divisions are all in the same orientation. I really think Canon needs to explore the idea of QPAF - quad-pixels (while a technical challenge, I think that would be than alternating/mixing the orientation of the pixel divisions).

I thought they were exploring and or working on this? I thought I read an article about it a couple months ago but I may be mistaken.

Sounds like an amazing advancement if they can make it so.....

 

[neuroanatomist]

That's the first I've seen a picture of a 1DXII PDAF sensor, I think.

That’s the 1Dx / 5D3 PDAF assembly.
The 1Dx II presumably has a different configuration of sensors to facilitate f/8 sensitivity at each location.

Correct. But they’re actually quite similar.

1D X AF sensor:

1D X II AF sensor:

 

[neuroanatomist]

Having never used DPAF, I don’t know how it works relative to orientation. I presume they construct lines analogous to dedicated sensors, not actually use a single photo site, but that’s a total guess. If so, what are the orientations of the lines, and does that affect their sensitivity?

If they are constrained directionally, that could explain the presence of selectable AF point which finds no subject.

The DPAF pixels are indeed constrained directionally, the hemi-pixel divisions are all in the same orientation. I really think Canon needs to explore the idea of QPAF - quad-pixels (while a technical challenge, I think that would be than alternating/mixing the orientation of the pixel divisions).

I thought they were exploring and or working on this? I thought I read an article about it a couple months ago but I may be mistaken.

Sounds like an amazing advancement if they can make it so.....

Good to know, thanks!

 

[3kramd5]

Having never used DPAF, I don’t know how it works relative to orientation. I presume they construct lines analogous to dedicated sensors, not actually use a single photo site, but that’s a total guess. If so, what are the orientations of the lines, and does that affect their sensitivity?

If they are constrained directionally, that could explain the presence of selectable AF point which finds no subject.

The DPAF pixels are indeed constrained directionally, the hemi-pixel divisions are all in the same orientation. I really think Canon needs to explore the idea of QPAF - quad-pixels (while a technical challenge, I think that would be than alternating/mixing the orientation of the pixel divisions).

Thanks. I suspected as much but didn’t want to state it as known fact.

I indicated in the battery charge precision thread that one of the things which this forum brings above others is technical detail. While it is sometimes misconstrued as a kneejerk defense of an alleged wrong, I see it as highly valuable. Understanding how your gear works is fundamentally important when you want to know why it isn’t working the way it is being used.

 

[fullstop]

DP-AF: "thanks to their split, each pixel serves as AF point" . and they all have the same orientation. fine. but why are there still only a relatively small number of "AF fields" on EOS M50 (99 respectively 143 with a few E-M lenses). are these "AF fields" then made up of a "strip" of Dual pixels? if yes, then why could these "AF strips" not be arranged as horizontal and vertical and/or diagonal crosses (as in the separate Phase AF sensors in DSLRs) to get best AF performance?

interestingly i cannot find any schematics or description of Canon DP-AF - neither in DSLR nor in EOS M that goes i to these details. seem to all stop at a more general level or descriptions referring to single "dual puxels". but not how the phase-AF system actually is set up. i conclude that there are no cross-AF fields, otherwise it would undoubtedly be promonently noted/marketed. so it must be "line sensors". will test whether i can only af on vertical contrast edges with M50 or also on horizontal structures.

 

[neuroanatomist]

DP-AF: "thanks to their split, each pixel serves as AF point" . and they all have the same orientation. fine. but why are there still only a relatively small number of "AF fields" on EOS M50 (99 respectively 143 with a few E-M lenses). are these "AF fields" then made up of a "strip" of Dual pixels? if yes, then why could these "AF strips" not be arranged as horizontal and vertical and/or diagonal crosses (as in the separate Phase AF sensors in DSLRs) to get best AF performance?

interestingly i cannot find any schematics or description of Canon DP-AF - neither in DSLR nor in EOS M that goes i to these details. seem to all stop at a more general level or descriptions referring to single "dual puxels". but not how the phase-AF system actually is set up. i conclude that there are no cross-AF fields, otherwise it would undoubtedly be promonently noted/marketed. so it must be "line sensors". will test whether i can only af on vertical contrast edges with M50 or also on horizontal structures.

This image show a little of the tech, notably that all the split pixels have the same orientation.

The above image is from a Canon DLC article on DPAF:

https://learn.usa.canon.com/resources/articles/2018/canon-dual-pixel-af.shtml

They could be sampled in any configuration, but still would be sensitive to phase differences in only one dimension (line sensors).

I really think Canon needs to explore the idea of QPAF - quad-pixels (while a technical challenge, I think that would be easier than alternating/mixing the orientation of the pixel divisions).

I was wrong on this, or at least Canon is pursuing the latter as a possible solution. This is from a Canon patent published late last year, showing a mix of pixel spit orientations:

 

[3kramd5]

DP-AF: "thanks to their split, each pixel serves as AF point" . and they all have the same orientation. fine. but why are there still only a relatively small number of "AF fields" on EOS M50 (99 respectively 143 with a few E-M lenses).

Notionally there are millions of combinations. Some considerations are:
1) How many pixels does it take to form an appropriate line? I assume given the split nature that a line pair isn’t required with DPAF like it is with traditional PDAF; one line is in essence a vertical pair (half the line looks left, half the line looks right).
2) Where are the appropriate locations for these lines?
3) Does the line configuration and/or location change based on lens settings?

Practicality limits the number of user-selectable locations. One wouldn’t want to scroll through millions of options. However it may be dynamic, not fixed, which is a unique advantage to using split pixels (smart Canon) rather than off sensor discrete lines and on sensor limited masked pixels.

 

[fullstop]

hmm, again: Canon says: EOS M50 = 99 AF fields / and 143 with 2 specific lenses. .
Canon also says "every DP-AF pixel on entire sensor serves as AF-field".

But: Canon does NOT say: 24 million AF fields.

So those 99 / 143 AF fields need to be some ARRAY of DP-pixels, right?
If so, does Canon arrange these "AF strips/lines" of DP-pixels in vertical, horizontal and/or diagonal cross pattern/s on sensor? if not, why not?

 

[3kramd5]

If so, does Canon arrange these "AF strips/lines" of DP-pixels in vertical, horizontal and/or diagonal cross pattern/s on sensor? if not, why not?

Presumably they only establish lines in the vertical. Why not horizontal or diagonal? I think I covered that in (1) above.

Each pixel has two photodiodes. One “looks left” and the other “looks right.” To be effective as a horizontal line sensor, they would have to “look” up and down. That could be accomplished with quads, or by alternating the orientation of the photodiodes within each pixel, but that would require a new sensor fab.

 

[neuroanatomist]

hmm, again: Canon says: EOS M50 = 99 AF fields / and 143 with 2 specific lenses. .
Canon also says "every DP-AF pixel on entire sensor serves as AF-field".

But: Canon does NOT say: 24 million AF fields.

So those 99 / 143 AF fields need to be some ARRAY of DP-pixels, right?
If so, does Canon arrange these "AF strips/lines" of DP-pixels in vertical, horizontal and/or diagonal cross pattern/s on sensor? if not, why not?

Probably squares. But, current DPAF is restricted to detecting features in only one orientation, determined by the split in the individual pixels. It doesn't matter whether the pixel sampling is in the pattern of a cross, square, or perambulating pentagon...the orientation sensitivity is fixed to that same orientation.

 

[fullstop]

i understand the "pixel level" orientation. But how does Canon arrange them to come up with 99 / 143 AF fields?
They must be "stripes/lines" of some sort. if they are horizontally aligned, looking left/right ... they should effectively phase AF fields that detect VERTICAL contrast edges, but "blind" tom horizontal lines, right?

 

[3kramd5]

i understand the "pixel level" orientation. But how does Canon arrange them to come up with 99 / 143 AF fields?

Nobody who contributes to this forum is likely to know that.

... detect VERTICAL contrast edges, but "blind" tom horizontal lines, right?

Edit: I misread the above.
Yes No, it’s the other way around. That is why I brought it up.

 

[fullstop]

there is some Canon "DP-AF voodoo" hidden. Why 99 fields, when they could just as well take [extremely spoken] 1 DP pixel towards left edge of sensor and 1 towards right edge, forming an (interrupted), extremely long "base-line" for Phase-AF? "99/143 AF fields number" really puzzles me. On a DP-AF sensor with every single pixel "AF-enabled".

???

 

[3kramd5]

Why 61 on the 1Dx?
Why 399 or whatever on the A7riii?

It’s almost certainly not arbitrary, but again nobody here is likely to have that detail.

Note it’s 99 user selectable points. The others pixels are likely capable of being used for tracking, like with the Nikon D5’s 153 points of which the user can only select about two thirds.

As for a single uninterrupted line, that wouldn’t really accomplish anything as far as I understand it, but I don’t claim expertise. The baseline has to do with the angles at which the sensors look towards the edges of the lens (hence different configurations being sensitive at different f-stops). Spanning the sensor wouldn’t change the baseline; it’s orthogonal to it. Neuro please correct me if I’m wrong and/or clarify.

 

[neuroanatomist]

Why 61 on the 1Dx?
Why 399 or whatever on the A7riii?

It’s almost certainly not arbitrary, but again nobody here is likely to have that detail.

Note it’s 99 user selectable points. The others pixels are likely capable of being used for tracking, like with the Nikon D5’s 153 points of which the user can only select about two thirds.

As for a single uninterrupted line, that wouldn’t really accomplish anything as far as I understand it, but I don’t claim expertise. The baseline has to do with the angles at which the sensors look towards the edges of the lens (hence different configurations being sensitive at different f-stops). Spanning the sensor wouldn’t change the baseline; it’s orthogonal to it. Neuro please correct me if I’m wrong and/or clarify.

Nailed it.

 

[rrcphoto]

hmm, again: Canon says: EOS M50 = 99 AF fields / and 143 with 2 specific lenses. .
Canon also says "every DP-AF pixel on entire sensor serves as AF-field".

But: Canon does NOT say: 24 million AF fields.

So those 99 / 143 AF fields need to be some ARRAY of DP-pixels, right?
If so, does Canon arrange these "AF strips/lines" of DP-pixels in vertical, horizontal and/or diagonal cross pattern/s on sensor? if not, why not?

it's 99 / 143 selectable points in some AF modes.

otherwise it's a freely moving box around 20 some odd million locations.

Canon has to do some seriously heavy lifting at times for AF on a DPAF sensor, my feeling it that each generation of DIGIC will see DPAF get quicker.

 

[pj1974]

it's 99 / 143 selectable points in some AF modes.

otherwise it's a freely moving box around 20 some odd million locations.

Canon has to do some seriously heavy lifting at times for AF on a DPAF sensor, my feeling it that each generation of DIGIC will see DPAF get quicker.

I have enjoyed following this thread (reading much of this thread has been interesting, not all of it though lol)
e.g. I found neuro's sharing about the mix of split pixel orientations particularly interesting!

Some time ago I was also thinking how much processing power DPAF takes. So, for argument's sake... (just keeping things simple, here folks)iIf the 'active AF square' represents ~5% of the sensor area of a 24 MP sensor, it equates to ~1.2 MP.

I wonder whether all of those 1.2 MP sensors are used, or an arrange (e.g. some along the horizontal, some along the vertical, and yes a spread across the AF square) - so it might be say 6,400 pixels (say 80 pixels x 80 pixels in the AF square) rather than the full 1,200,000 pixels.

Particularly because AF is making lots of calculations from these (sampled?) AF points, as it moves from out of focus to in focus. My IT understanding is decent, however others would have better know-how as to how many calculations it takes (per second / millisecond) to accomplish one simple 'DPAF' out of focus to in focus adjustment.

Then where continual AF tracking (both for live-view stills and video) comes into play, even more processing is required. So I agree with rrcphotos' post above, that with increasing Digic processing power (and potentially efficiency, i.e. # of cores, etc) will see DPAF speeds increase in the future, perhaps at exponential increments when new processing technologies are implemented. 8)

One aspect I would like to see in improved future DPAF technology, would yes, potentially be QPAF, and improved ability / sensitivity in low light. (e.g. reliable, fast AF down to -4EV would be lovely!)

Regards,

PJ