Surprisingly narrow "Depth-Of-Field"

[CanadaDrone]

I was under the impression PDAF really needs a mirror to work well.

It actually does not - on sensor PDAF (OSPDAF) has been around for a long time now and works in a quite similar way to traditional PDAF. The first really successful consumer iteration was in the Nikon 1 V1 camera in 2011. After that, Sony and Fuji got on board pretty quickly, and now just about everyone has it (most implementations using Sony sensors). Just like DSLRs, not all OSPDAF systems are created equal, but today, the very best systems from Sony, Nikon, and Fuji are only slightly behind the high end DSLRs and perform very well.

In the simplest terms, most of these OSPDAF systems work in basically the same way as a traditional PDAF array by comparing two lines of pixels masked to look at light from two different areas through the lens which allows it to calculate a phase difference. In cameras that allow you to select different focus points, you are basically selecting left/right pairs.

Generally OSPDAF struggles in two areas - discretion/geometry and low light capability. For the discretion issue, the greater physical distances afforded by the geometry of the traditional PDAF system (sensor in the bottom of the camera using a sub-mirror to see through the lens) allows for more accurate results particularly at longer focal lengths (300mm++). The downside of this method is that manufacturing and adjustment tolerances can come into play as they are set in micrometers. Most high-end OSPAF systems now too have improved their low light performance, but they are still slightly behind the very best DSLR PDAF systems (in really low light they switch to CDAF).

The other way cameras with OSPDAF get around the discretion issue is by forcing a CDAF confirmation step at the very end of the PDAF cycle - this makes it very fast and very accurate for stationary subjects. The trouble is when you switch over to continuous autofocus, every current camera with OSPDAF currently foregoes that final CDAF confirmation step at the end of the PDAF cycle, sacrificing a tiny bit of critical AF accuracy for speed. Generally this isn't a problem, especially with extra DOF offered by smaller apertures, but it is noticeable at wider apertures. I would say the very best OSPDAF systems like that in the Sony A9 are about 80% as good as the very best PDAF systems like those in the Nikon D5/D850/D500, and that is being quite picky. In the future, there might be enough processing power to brute force the CDAF confirmation step at the end of the fastest AF cycles (i.e. 20FPS), but right now it is not possible.

So in summary, the best OSPDAF systems are only slightly behind the best PDAF systems, and work in a very similar way, but both have a wide range of performance and generally you get what you pay for. I highly doubt the OSPDAF on the M2Zoom is very good, but it is almost certainly miles ahead of any CDAF-only system like that used on the M2P for anything other than stationary shooting.

Regarding "infinity focus" - how accurate is the calibration of the "infinity focus" setting? I remember there had been issues in that arena on the Mavic 1.

All I can offer here is that I haven't read of a single person having an issue with it yet on the M2P. At the distances of the typical drone footage, and the massive DOF awarded by the tiny 1" sensor, it is going to be a non-issue the majority of the time. I think you'd find out pretty quick if there was an issue.

 

[Roamer105]

CanadaDrone - thank you for your detailed explanation.

 

[stevenws]

The other day I shot some video from take-off without specifically focusing the camera. I was surprised at this footage shot in a way that demonstrates how narrow a depth-of-field this particular lens has. I don't remember what the aperture setting was but I am sure it was less than 5 - most likely at or close to wide open. Nevertheless, clearly, focusing correctly is quite critical with this lens.

View attachment 50565

I'm not happy with my mavic pro camera because of this. Seems the camera is for still portrait photography instead of beautiful landscape videos. Why would DJI put this camera on a drone is beyond me!?

 

[stevenws]

Now look at anything you've shot from up in the air where your subject is more than 10 metres away from the lens (like almost every photo ever taken with a drone) and you see that the lens has more depth of field than you can use.

What you've observed is the small zone of focus when the lens is at the extreme close end of its focus range.
I would expect that the number of photos taken with a drone within 2 feet of the subject to be something like 1 in a million.

So, even though the apeture is fixed at 2.2, you are saying this changes with height? To me, being high just means this extremely shallow DoF is just not as noticable... but still there?

 

[Meta4]

I'm not happy with my mavic pro camera because of this. Seems the camera is for still portrait photography instead of beautiful landscape videos. Why would DJI put this camera on a drone is beyond me!?

What the OP said in his first post was incorrect.
The lens does not have a narrow depth of field at all, in fact is has more depth of field than you will ever need.
That also applies to the original Mavic pro.
See my explanation in post #3 & #6.

So, even though the apeture is fixed at 2.2, you are saying this changes with height? To me, being high just means this extremely shallow DoF is just not as noticeable... but still there?

The height of the drone makes no difference.
If the lens is focused correctly, the distance of the subject from the lens makes the difference.
And one more time ... the lens has lots of DoF rather than a narrow DoF.

 

[stevenws]

I have had a Canon D6 for about 6 yrs. Even do some hobby portraiture with it. Sorry, but your explanation makes no sense to me. Thanks for your input.