contrast based af vs phase detect af in real world

[zlfan]

 

[Davide DB]

I am beginning to develop an allergy and even hatred toward everything I see on YouTube. What a bullshit title. People who have never filmed shit other than cats or begonias in the backyard.

I, with my GH5, am delighted with its performance underwater and on land.

 

 

[zlfan]

 

s5's video af seems not far behind the s5ii. 

 

[zlfan]

I think that in the scenarios of complicated movements by many people in low light, phase detection af may be more reliable. for many common shooting conditions, the contrast based af may be ok. 

phase detection af uses pixels for focusing, so the image quality is actually a little bit lower than contrast based af. each technology platform has its own pros and cons. just prioritize the needs then choose the right technology platform. 

[MrSMW]

7 minutes ago, zlfan said:

s5's video af seems not far behind the s5ii

Hilarious.

The horse & cart is not that far behind the motor car.

OK, the S5 is not quite as bad as some make out but the single biggest issue between these two and tests is as follows.

The S5 and the S1/S1H/S1R, are finicky bastards when it comes to video as in there is a combination of settings and tweaks that need to be made to achieve the optimal AF within their capabilities.

Then there are circumstances to add into the mix.

And then lens choices as in real native as in Lumix, or L Mount native such as Sigma.

Then differences between zooms and primes.

Flip to the S5ii and 90% of these issues cease to exist.

Of course, same as with any camera, AF can be optimised and optimised for certain specific scenarios, but the S5ii is way ahead of the S5 regardless of what some random YouTuber has to say.

And that is not to denigrate this particular YouTuber as I have no idea who he is and nor have I watched the video, - just speaking from experience.

I have not tweaked a single setting since moving to S5ii's from S5.

[Davide DB]

The scenarios can be very different.
If I have the camera on a gimbal I have always tried to be hyperfocal, never attempted to use CAF with the cameras I have had (all contrast based). There is always a moment when you can lose focus and you are screwed. Maybe with PDAF it's different and you can be safe. In other more static scenarios I have always used single AF and then worked manually with focus peaking. If you then have a manual lens with a nice soft dial, it's fun and just takes a bit of practice. In fact I hate Lumix lenses with focus by wire. Only in the latest lumix cameras and only with some lumix lenses you can set up linear rotation otherwise it's a nightmare.
Underwater you work hyperfocal as much as possible or in manual via focus peaking. But the cases in which it is possible to do this are very few. Only cinema cameras and cinema housings allow you to work with manual focus reliably, but we are talking about kits that exceed 20K euros. Normally the focus ring on the housing demultiplies the revolutions of the lens focus ring. You turn the focus ring three times and the lens moves the focus by 1 cm. impossible to use it outside macro shooting.

[zlfan]

19 minutes ago, Davide DB said:

The scenarios can be very different.
If I have the camera on a gimbal I have always tried to be hyperfocal, never attempted to use CAF with the cameras I have had (all contrast based). There is always a moment when you can lose focus and you are screwed. Maybe with PDAF it's different and you can be safe. In other more static scenarios I have always used single AF and then worked manually with focus peaking. If you then have a manual lens with a nice soft dial, it's fun and just takes a bit of practice. In fact I hate Lumix lenses with focus by wire. Only in the latest lumix cameras and only with some lumix lenses you can set up linear rotation otherwise it's a nightmare.
Underwater you work hyperfocal as much as possible or in manual via focus peaking. But the cases in which it is possible to do this are very few. Only cinema cameras and cinema housings allow you to work with manual focus reliably, but we are talking about kits that exceed 20K euros. Normally the focus ring on the housing demultiplies the revolutions of the lens focus ring. You turn the focus ring three times and the lens moves the focus by 1 cm. impossible to use it outside macro shooting.

all true. hyperfocal and zone focus etc are really helpful. 

I shoot photos using olympus e-1 with shg 14-35 f2 and 35-100 f2 and 55-200 f2.8 many times, as I like the e-1's ccd. e-1's af is dinosaur according to current standards, but I never feel the lacking of a modern af. I shoot outdoor daylight events most of the times, lighting is not an issue. I use single af plus mf. I used to use f2, but quickly found out that the editors don't like the blurred background, as they want to see a group of people gathering and all of them need to have in focus face. I later use f4 all the time, much higher keeper rate.

my point is that not every scenario needs the latest technology platforms, as you move on to these newer platforms, something else may be lost. it is always balancing. 

I totally understand that for indoor sports, low light wedding, events like these need the best af for photo, and c-af for video. 

[Davide DB]

One last consideration for underwater use and then I won't bother with underwater any more.
I am very curious to try continuous autofocus with the PDAF in water. All these videos always refer to Face Detect (and cats dogs birds). The algorithms are trained on these subjects by machine learning. What happens with fish? There are plenty of videos on Sony, Canon and Nikon in which they use the animal eye AF function without any problems and it seems to work perfectly. but these are all photographic tests. I'm very curious to try it out on video.

[zlfan]

43 minutes ago, MrSMW said:

Hilarious.

The horse & cart is not that far behind the motor car.

OK, the S5 is not quite as bad as some make out but the single biggest issue between these two and tests is as follows.

The S5 and the S1/S1H/S1R, are finicky bastards when it comes to video as in there is a combination of settings and tweaks that need to be made to achieve the optimal AF within their capabilities.

Then there are circumstances to add into the mix.

And then lens choices as in real native as in Lumix, or L Mount native such as Sigma.

Then differences between zooms and primes.

Flip to the S5ii and 90% of these issues cease to exist.

Of course, same as with any camera, AF can be optimised and optimised for certain specific scenarios, but the S5ii is way ahead of the S5 regardless of what some random YouTuber has to say.

And that is not to denigrate this particular YouTuber as I have no idea who he is and nor have I watched the video, - just speaking from experience.

I have not tweaked a single setting since moving to S5ii's from S5.

I respect and believe your hands on experience. 

in the first video, the guy said that he has shot 500 weddings during the last 10 years. he used this s5 for several years, and did not feel the lacking of s5's af. I was impressed by his user experience so shared here. no offense. 

[MrSMW]

1 hour ago, zlfan said:

no offense

None taken, none given, - this is all just chat/opinion!

[Al Dolega]

Remember that "phase detect AF" systems are also using contrast detect in combination with their phase system. So really the comparison is phase+contrast vs. contrast only.

My S1 was always significantly better in 60p than slower framerates, which makes sense, more samples per second, more chances to check and compare focus. And when it worked the performance was pretty good! But the inconsistency/unpredictability, when it would wander off on its own seemingly at random, was more frustrating than any slight performance deficiency it had when it did work.

My S5iiX is night and day better, still not perfect of course. I don't think any AF system will ever be perfect, no matter how much processing power and AI they throw at it, at least not without some form of input or guidance from the user.

[MrSMW]

2 hours ago, Al Dolega said:

And when it worked the performance was pretty good! But the inconsistency/unpredictability, when it would wander off on its own seemingly at random, was more frustrating than any slight performance deficiency it had when it did work

That was the sum of it for me and many others.

And for my work, I need reliability above all else because even an average result is infinitely better than no result at all.

[zlfan]

seems to me that s 5 III is worthy of waiting. very possibly it will have internal prores raw codecs, improved ibis, better af. 

[newfoundmass]

12 hours ago, zlfan said:

 

s5's video af seems not far behind the s5ii. 

 

Having owned both, the S5ii is much more reliable. But the AF "issues" with Panasonic cameras were always overblown in most instances and easy to manage. The biggest damage the dfd system did was to their reputation. I still use two S5s to film pro-wrestling, a pretty unpredictable endeavor, and the AF is fine, which if you'd only seen unpractical AF tests from Youtubers you'd never believe was possible  

 

It IS though almost absolutely unusable for small objects and animals in video

[Al Dolega]

7 hours ago, zlfan said:

seems to me that s 5 III is worthy of waiting. very possibly it will have internal prores raw codecs, improved ibis, better af. 

I think you'd be waiting for at least a year. The S1Hii seems to be set for later this year or early next year, possibly an S1 or S1R replacement as well. They wouldn't do a new S5 until well after those release, and they wouldn't be able to get away with doing another IMX410 iteration, so it would probably depend on an better-enough sensor getting cheap enough to keep the S5 at the same price. The Z6iii sensor is probably exclusive to Nikon for a while, and then Sony might use it, so that might be out of the picture for a year or two.

[newfoundmass]

On 7/3/2024 at 9:40 PM, Al Dolega said:

I think you'd be waiting for at least a year. The S1Hii seems to be set for later this year or early next year, possibly an S1 or S1R replacement as well. They wouldn't do a new S5 until well after those release, and they wouldn't be able to get away with doing another IMX410 iteration, so it would probably depend on an better-enough sensor getting cheap enough to keep the S5 at the same price. The Z6iii sensor is probably exclusive to Nikon for a while, and then Sony might use it, so that might be out of the picture for a year or two.

Yeah, I wouldn't expect there to be another S5 until 2026 sometime. There really isn't a reason to release a mark III sooner than that, especially with whatever else they have in the pipeline. Even a year from now the S5 II/X will be a really good value, and they can lower the price if needed to keep it competitive. They hit a home run with their entry model, now it's time to focus on their other cameras.

[zlfan]

From a pure lab test perspective, it seems that there is no gain without pain: the new LUMIX S5 II has nice new features including solid autofocus, but unfortunately in the lab, it falls a bit behind the previous generation, the S5 – when pushed to the absolute extreme, to keep things in perspective. 

Not only is the rolling shutter worse (1ms more in full frame, 1.8ms more in APS-C), also the latitude result is not as good – it finishes with 7 stops of exposure latitude, mainly because of banding artifacts in the shadows and a more blotchy noise pattern which doesn’t look as pleasing to the eye.

That said, the dynamic range results are very similar to the previous generation LUMIX S5.

 

from cined test.

 

[kye]

I love these discussions - it reminds me how people have almost no understanding of how anything works.

CDAF vs PDAF has almost nothing to do with the AF performance of modern cameras.

[BTM_Pix]

I used the Animal Eye AF on my S5ii the other day for the first time and whilst it was reliable in stills, in video it was certainly less so as these frame extracts show.

When it worked, it worked well.

But in a lot of cases it preferred his nose.

This was with the Panasonic 85mm f1.8 so there isn't a massive margin for error but it also has to be said that Rolo is what can politely be described as a lazy little fucker so he's not exactly doing a lot of leaping about to overly tax the detection and tracking.

Maybe it thought his nose was a built in Owl and it was picking up its eyes.

[Davide DB]

Yet there are hundreds of videos showing how animal eye AF on land works smoothly on Canon and Sony cameras. But this has nothing to do with CDAF and PDAF, it depends on the data the algorithm was trained with.
AFAIK algorithm deep learning data are embedded in HW so it is not possible to improve recognition via firmware update.