DPReview TV: Do the Panasonic S5's updated subject recognition and DFD improve autofocus in video?

To be fair any cameras released in 2020 should have reliable autofocus. This goes for the R5, A7siii and any upcoming cameras.

If Panasonic's cameras were designed for MF, why bother developing the DFD AF system at all? They could have settled for conventional contrast detect AF, or no AF at all. In no way are Panasonic marketing their cameras specifically for MF users, on the contrary, they have stated that they think DFD is preferable over PDAF.

"Panasonic's cameras are designed for MF users. IF you want reliable AF, no stupid buy Panasonic"

Panasonic AF is reliable......I proved it just now using AF on a tripod focusing on a brick wall.

@Ihkjacky, yes specially designed for MF, but they don't include punch in to focus functionality while recording.. great MF design.....

Left eye, it is definitely not as good as the Sony, or the Canon. Not a single review says that it is.

The S5 comes with head detection out of the box, and the S1/R/H each recently got it through a firmware update. The whole point of it is to help the cameras keep focus locked on the subject even when they turn their face away. It's not a CDAF vs. PDAF thing, it's an AF algorithm thing. That said, I personally would not consider buying any of Panasonic's cameras due entirely to their lack of PDAF.

I don’t remember which one but one reviewer mentioned it has timecode.

That being said, the S-lineup will get an upgrade with S5 AF.

It wouldn’t surprise me as Sony is always seeming to push the technology envelope. At the same time it doesn’t matter to me whether Sony is working on TOF. Their system is rock solid already. Although TOF systems may be able to improve distance detection rates at sampling frequencies much greater than 60 per second. It may be more useful for them in their RX series and cell phone applications also.

I suspect it’s your #2. For any kind of motion blur the camera really has no way to know if the softness is due to defocus or motion. So how can it know whether it’s a focus problem or not? Much less so which way to go?

I think these are just the inherent limitations of a system that doesn’t have any kind of distance sensing (like PDAF). I don’t think it would be possible for Panasonic to catch up to the performance of the latest on sensor PDAF implementations without some PDAF or other distance detection capability.

My understanding is that it's not so much shutter speed that improves the AF performance, but rather frame rate, since the system has more samples to work from.

So 60 fps is better than 30 fps is better than 24 fps.

Seems like the frame rate vs. shutter speed distinction is strange thing for an expert like Jordan to conflate.

Sure. If it’s extremely blurred yes. And PDAF knows more than just the direction. It knows the magnitude of the phase difference which is a proxy for how much the focus needs to shift.

If you look at AF performance overall with on sensor PDAF systems from Canon, Nikon, and Sony, I don’t believe the on sensor systems are inferior in capability to off sensor dedicated PDAF systems.

OSPDAF performance is somewhat limited at longer focal lengths vs traditional PDAF due to the geometry of the phase sensors. Even for nominal focal lengths the half-masked OSPDAF designs design's can't optimally handle the range of exit pupils encountered. They work around this by using varying masked sizes across the PDAF array, biasing to the ones that closest match the exit pupil of the lens mounted. You can read about this in Sony's patents for example, this one: https://www.freepatentsonline.com/20110063484.pdf - start reading at [0081]

For professional work there is no camera that is 100% reliable. If there is more than one subject no system picks the right one 100% of the time. And no camera does focus pulls exactly the way a film maker wants 100% of the time. And like you said 98% is not good enough,
If you are serious about making videos you learn how to manually focus, and use it when focus is critical.
Also, for vlogging while the S5 has issues, it is good enough for most. I've seen a lot of missed focus in vlogs where no one cared. The Engadget reviews called the S5 the best $2000 camera for vlogging, and they may be right.

@left eye @Noah M

Well, your 100% figure is wrong, as I know quite a few professionals that rely on autofocus. One man crews that rely on gimbals, like many wedding videographers have done so for years.
And while for the rest autofocus might be new there is no question it will take over the industry.

Yes, a few years ago the industry was 100% manual focus. But that wasn't because AF was not wanted, that was because nobody had reliable enough AF. Now with AF getting good you will see quite a quick shift. There is a reason both Canon and Sony now also build good AF into their dedicated C and FS video line

panther fan,
"Well, your 100% figure is wrong, as I know quite a few professionals that rely on autofocus."

that doesn't mean AF is 100% perfect. And its not. If you want to embarrass yourself and claim AF in some cameras is 100% perfect, go ahead. And if you want embarrass yourself and claim that some cameras do focus pulls exactly like film makers want 100% of the time, go ahead.

@left eye But how many *would* use AF if they had a reliable system? There was a time when pro stills photographers all used manual AF, too, because stills AF systems were simply too primitive. Over time, stills AF became both faster and more reliable than manual AF, and at that point, pros mostly switched to AF. I believe the same will happen with video. The great video AF we've found on the R6/R5/a7S3 opens up so many possibilities not just for vloggers but for videographers, filmmakers, reality TV, documentaries, etc. They can now use shallow DoF, tracking shots, etc., with minimal setup. It really has amazing potential and I believe it will change the look of video over the next decade.

I think Panasonic is betting on TOF. But I can't say if it will do what they hope.
TOP in combination with DFD would might be the best solution. PDAF blocks a small number of pixels and in rare cases causes PDAF striping.

@Noah M
I am sorry but have you read any TOF papers? That long-distance Panasonic sensor uses 1.2kW of power. Meaning your 14Wh battery will last you a cool 40sec, while at the same time you grill your model

ToF has numerous problems besides that, like the fact that most sensor placements on the camera will be blocked by the lens. Or that you would actually need very precise lens calibration to make use of it.
Currently, Sony is the most advanced manufacturer of ToF sensors. Both for iToF phones as well as dToF like in the apple lidar system. If anyone could pull it of it would be Sony. But they won't because it's a bad idea.

panther fan, You see to not be able to read. I only said Panasonic might be betting on it, but "I can't say if it will do what they hope."

But then you are the guy who swears AF is always 100% perfect during video with some cameras, So we don't expect you to understand,

@panther fan - to be fair, if I am reading the paper correctly, the pulse power of the laser is 1.2 kW, but that pulse is only 10 nanoseconds long, so each pulse only uses 3.33 × 10^-9 watt hours. Even at 6 kHz, that's a continuous energy consumption of 0.07 watts. Near negligible, at least for the laser alone. That seems too low, but 1.2 kW continuous is far, far too high.

Furthermore, that is for a LiDAR system with a ranging capability of 250m, which is clearly far longer than is required for an autofocus system. Even a wide aperture exotic supertelephoto is hyperfocal after 100m. A much shorter range sensor - good to 10m, for instance - would use much less power.

It looks like shorter range sensors from Broadcom (5-100m, 3 kHz) have a typical power consumption of 0.17W (33 mA * 5V) only.