Determinants of autofocus function

[sherman_levine]

At equal sensor technology an MFT sensor will have 2 stops more noise than a FF sensor, which means worse phase detection in marginal lighting situations, the result being more hunting and potentially less accuracy, the latter depending on how much time the AF system allows itself to focus in a given cycle before letting an exposure to go off.

Adam, I understood the OP as requiring equivalent setups, which imply a similar number of captured photons, which imply similar SNRs at the same output size. Where do the 2 stops of additional noise come from?

Yep, I saw that too Jack but wasn't sure if that's what the poster meant by it.

If that is what he meant it's not clear to me how you get equivalency if both AF systems focus at the lens's largest aperture (independent of shooting aperture), which presumably is the same between lenses in this theoretical setup.

I'd seen the statement that when FF and MFT cameras were set to create equivalent images the MFT camera would focus more quickly because the exposure was greater - but I was not able to find any data to support that statement.

Beyond that I was interested in learning how phase detection sites and imaging sites shared the same sensor. Did a phase detection site substitute for an imaging site (in which case its size would vary with pixel count) or was it something entirely independent.

Thank you

The PDAF pixels are of the same type as the imaging pixels. The difference is the PDAF pixels have a metal mask in front of them that blocks light from half the angles. The pixels are in pairs so that alternate halves are blocked, with two pixels making up a single PDAF sample point. Typically the AF sensors you see depicted in the viewfinder/LCD are composed of a set of these pairs. Lastly, the PDAF pixels have a fixed CFA type in front of them...green if my memory serves correctly.

The metal masks size makes the pixel sensitive to a range of focal lengths / exit-pupil distances (ie, angles of light), so to handle the full range of expected focal lengths there are usually a variety of metal mask sizes distributed throughout the sensor.

Since the PDAF pixels are of the same type as the imaging pixels they share the same noise characteristics, so your instinct about pixel size playing a role in the PDAF system sensitivity is correct.

Also, since PDAF pixels occupy a pixel that would otherwise be an imaging pixel but do not themselves contribute to the image, the values in the final image for their positions is interpolated from the imaging pixels around them.

Thank you for the information. I did find a few links which might be of general interest.

Describes (among other things) heterogeneity among sensor designs

https://www.sansmirror.com/articles/choosing-a-mirrorless-camer/autofocus-systems.html

Some discussion from DPR forums which includes links

https://www.dpreview.com/forums/thread/4703538

--

Sherm

Sherms flickr page

P950 album

P900 album RX10iv album
OM1.2 150-600 album

 

[sherman_levine]

I'd seen the statement that when FF and MFT cameras were set to create equivalent images the MFT camera would focus more quickly because the exposure was greater - but I was not able to find any data to support that statement.

You are correct about exposure but that doesn't automatically translate into better performance:

Equivalence requires the same number of pixels, so FF's would be proportionately larger, and the number of photons captured by each would be the same (because of throuput).

If you allowed for more pixels in FF, then they could have more OSPDAF ones, and they do, which could compensate for the different exposure in a number of ways.

In the end the proof of the pudding is in the eating and depends on each camera's specific variables.

Jack

Thank you. Took me a bit to recognize that the focus sites were repurposed/multipurpose photosites. After that, it all came together more clearly.

--

Sherm

Sherms flickr page

P950 album

P900 album RX10iv album
OM1.2 150-600 album

 

[OpticsEngineer]

"the proof of the pudding is in the eating and depends on each camera's specific variables."

Just for fun just now I closed the blinds in the living room to get a low light environment. Then I tried out the AF on my full frame Canon RP with RF 24-105 lens. And on my MFT Olympus OM-1 with Olympus 12-40 lens. I set the zoom to about mid range

Both cameras have on sensor PDAF.

First I focus on a chair about three feet away, then on some orange shoes about fifteen feet away.

The image snapped into focus on my Olympus very fast. I can't estimate a time that short. The image on the Canon gradually came into focus. Took about a second. Really irritating. I had forgotten why I hardly ever use that camera.

The image also snaps into focus very fast with my older Olympus EM1.2 with the old version Olympus 14-150.

I tried various modes, P, A and Auto on the RP and EM1 (the OM1 doesn't have Auto) Various AF modes. Result was always the same.

Of course, the RP is the cheapest Canon RF camera. And the Olympus models are top of the line. So as you observed, a lot depends on the camera and the lenses.

 

[Richard Butler]

I think a factor that's being talked around but not spelled out is that cameras don't necessarily focus using the aperture value that they then expose at.

So, even if the exposure of a Micro Four Thirds camera and a full-frame one were set to equivalent values, that doesn't necessarily mean that the full-frame camera will stop-down to the shooting aperture as it focuses. So unless the maximum apertures of the lenses are equivalent, you can't assume that the cameras will focus using equivalent apertures.

My worry is that the statement being tested sounds a lot like a hypothesized advantage of Micro Four Thirds, re-stated as a definitive benefit. I'd be wary of it without extensive (multiple brand) real-world testing.

Richard - DPReview.com

 

[Doppler9000]

I think a factor that's being talked around but not spelled out is that cameras don't necessarily focus using the aperture value that they then expose at.

Couldn’t changing the aperture cause focus shift?

 

[Richard Butler]

I think a factor that's being talked around but not spelled out is that cameras don't necessarily focus using the aperture value that they then expose at.

Couldn’t changing the aperture cause focus shift?

Yes, which is presumably why the exact approach varies between brands (I think Sony lets you choose between two modes, if you delve deep enough into the menu). I'd expect most systems to make some sort of correction for anticipated focus shift.

Richard - DPReview.com

 

[OpticsEngineer]

I think the trend of the discussion up to the point where I did a test was that full frame would always focus faster and more accurately then MFT because of the wider baseline for PDAF. And there was also confusion from old articles about PDAF that pre-dated on sensor PDAF.

So I just wanted to do a quick check to see if the trend of the discussion would bear up to an actual test, which it did not.

Of course my better full frame cameras and lenses focus as fast or faster than my best MFT as near as I can tell.

 

[J A C S]

Also, better or not at what apertures? 35/1.4, 50/1.2 or 135/1.8 wide open? M43 cannot do equivalents of those in the first place. 25-105/4 or 24-70/2.8? Same thing. When you stop down, accuracy is not a problem, not that it is wide open, either. The question might be meaningful for longer lenses with some popular almost equivalent lenses like 300/4 on m43 and 100-500/4.5-7.1 on Canon Ff at the long end because then the practical limits are determined by price, size and weight.

 

[Bill Ferris]

And as previously mentioned in the thread, there are factors determining focus accuracy that are not driven primarily by light-gathering. Consider subject detect autofocus. The driving force there is how capable the machine learned algorithm is at identifying a bird or other animal in the frame. Based on the performance users of all brand cameras report, it's clear this tech is in its infancy.