The upside to A-mount popularity going down

Started 3 months ago | Discussions thread
ProfHankD
OP ProfHankD Veteran Member • Posts: 7,861
Re: Some details about focus micro adjustment

SQLGuy wrote:

ProfHankD wrote:

I vaguely recall it not being too terrible to mechanically adjust the PDAF sensor alignment on some DSLRs, but you had to go in from the bottom, and it wasn't a user-friendly thing to do.

I think there are two different things going on here.

One is overall adjustment of the PDAF sensor system to correct differences in light path distance to the PDAF array and to the sensor plane. This is adjustable in most cameras by shims (Canon calls them washers) and/or by other mechanical adjustments (Nikon takes a small Allen key in the side of the mirror box to adjust this offset).

That's the biggie.

The second is, I think, a difference in how red/infrared vs mid-band visible light is focused through a particular lens. That's my theory, at least, of why it's different from lens to lens. In practice, I can guarantee you that it is. For instance, on my A900, the Tamron 70-300 USD gets -14 while the Minolta 100-300 APO gets +8.

It is possible that you'd need to correct if it's using wavelengths that far out for PDAF, but I don't think it is. If your PDAF is really correctly in line with the sensor, you should NOT need any correction unless you're using things like NIR filters. It sounds like your A900 is simply out of alignment, and the A900 is a traditional SLR, so it does have that extra complex PDAF optical path... and it also has the micro adjust support so misalignment can be claimed to be your problem, not Sony's.

... The SLT design makes precise alignment of the PDAF sensor an order of magnitude easier and easily much more accurate: it basically sits where the OVF focus screen would be, with light split to it by the non-moving pellicle mirror (what SLT really should have been called).

Maybe they didn't want to confuse things with the Canon Pellix.

Clearly, Sony didn't want to appear to be copying an old Canon. However, the term "Translucent" is technically (mildly) incorrect. I always figured the term was invented by a non-native English speaker placed high enough in the company so that nobody would be so impolite as to correct them... very Japanese.

My question is: why didn't they implement automatic MFA on SLT bodies? They could fairly easily have used CDAF as the control in a loop with MFA adjustments to minimize the disagreement between where the PDAF module saw focus and where the sensor saw focus.

Great question. Heck, the noise-cancelling Sony headphones I bought recently not only autocalibrate for fit around your ears, but also can be tuned to the user's ear shape if you send Sony photos of your ears so they can make a 3D model. It would be very easy to autocalibrate -- especially using the LA-EA2/4 adapters with main-sensor PDAF able to be used to calibrate without even needing a lot of different focus distances.

I can only say it obviously never was a priority.

After all, remember that, except for a few stragglers, all Sony A mount lenses have had built-in focus motors. We're really talking about support of Minolta legacy glass -- and in some cases, 3rd-party glass.

Off-sensor AF still has the advantage of being able to focus in the dark with an IR assist pattern.

IR? Well, that's a problem because NIR focus plane rarely matches visible. BTW, nothing prevents use of focus assist lights with main sensor AF.

It also, I think, still has advantages for cross-type sensels.

In Sony's implementation, it does. However, there's no reason masked or dual pixels (really the degenerate case of plenoptics) can't support various different angles, and some such sensors have been built.

Bit by bit, though, on-sensor AF is meeting, and largely, exceeding, off-sensor AF.

Actually, the big disadvantages using the main sensor are the (usually negligible, but not always.. and yeah, I'm the guy who created that fix ) flaws they introduce in the image (e.g., similar to some unfortunately-correlated bad pixels for masked pixels) and the less-selective angle sensitivity. However, you can have so many more PD points on sensor that sloppier processing actually returns higher-quality phase information. In fact, Sony's original patent on their masked-pixel structure suggests an amazingly dumb alignment computation... which works shockingly well.

You'd think dual pixels (as used by Canon, and also by Sony in their cell-phone sensors) would work better because it gives even more data points, but at least the Canon dual pixels are less selective and "leaky" enough (as I measured with the 5D IV dual-pixel raws) that they actually produce significantly poorer phase information without a fancier algorithm... which is probably why it took a while for Canon's dual pixels to approach the speed of Sony's masked pixel AF.

Also, Sony doesn't just do PD on the main sensor, but combines it as a hybrid with CD, which is a very smart pairing.

Honestly, I think the correct longer-term answer is depth from defocus (DFD), which doesn't require any special pixels -- just fast readout. The DFD in MFT bodies is multi-shot DFD, using multiple shots to avoid much more complex computation, but single-shot DFD is feasible with accuracy better than PDAF. Again, the Sony A1 sensor tech opens the door to a lot of potential improvements -- not just eliminating the mechanical shutter, but also things like moving to a really great DFD implementation that has no artifacting in the image.

 ProfHankD's gear list:ProfHankD's gear list
Canon PowerShot SX530 Olympus TG-860 Sony a7R II Canon EOS 5D Mark IV Sony a6500 +30 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow