EOS RP front focus with EF lenses - Please CANON add MFA to RP!

koenkooi wrote:

Peak freak wrote:

I've been thinking! (I've been warned about that, but I still do it...)

My (recent) understanding was that the sensor acquires focus and drives the lens to the required position. This may allow for lens errors to show up.

However, the sensor 'sees' through the lens. It knows focus only when the lens gives optimum contrast or objects are in phase. In other words, the sensor (or camera) doesn't tell the lens were to focus. It just knows when things are in focus because it sees that they are. If the lens isn't focused properly then the sensor won't get correct contrast or phase - it should keep moving the lens focus elements until things line up.

So, mirrorless on sensor focus should be infallible. What am I missing? Are tolerances (probably for speed) loose for older lenses? Why though could the RF70-200 front focus?

You're describing a closed-loop system, which for RF lenses on R bodies doesn't seem to be the case, DPAF figures out the distance and then tells the lens to move there. The RF70-200 issue seems to prove that theory. It's very fast, but depends on lenses doing the right thing.

For EF lenses on R bodies it seems to use something more like a closed loop, but no contrast based finetuning at the end of the focus cycle, it's all DPAF. Using a 26 year old design like the 180mm L shows a low frequency stutter when focussing, not a high frequency one you'd expect with a naive closed loop algorithm.

This is all conjecture on my side and my RP isn't the best example of R-series AF

I wish Canon would offer the options for a) doing closed loop focus and b) using contrast for finetuning the focus lock. The last one is handy when your subject has the most contrast in the wrong orientation for DPAF.

And for bonus points: export a depth-map with each picture, the information is already there since every pixel is a DPAF pixel.

"DPAF figures out the distance". Do phase detect pixels only need to know how out-of-focus a subject is? [Then know where to drive the lens to get the subject in focus without actually confirming that it is in-phase.]

My assumption was that phase detect relies on moving the lens focus elements until the subject is in-phase (and in focus). Maybe it only needs to know how far OOF it is. Quite plausible.

Might have a bit of  read up. Panasonic uses that 'Depth from De-focus'. Might explain a few things.