I know Sony disables PDAF in lower light, and Olympus may too. There are only about 100 pixels out of millions and any noise might confuse PDAF. DFD uses more pixels and while there will still be noise it should be much less of an issue an no need to turn it off.
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
GH4 focusing working well so far.
- Thread starter Lab D
- Start date
True, although I imagine that the statistical analysis of blur characteristics will also get less reliable as noise levels go up. We'll have to wait for more user experience with the GH4, and then for the next iterations of the technology. CDAF took a long while to get this good, and for years the consensus was that it could never be anywhere near as fast as PDAF.
Adding: that problem with on-sensor PDAF may also be solved with Canon's approach of making every pixel on the sensor a split pixel.
Last edited:
I agree that more testing needs to be done to see how well DFD works, but it's not accurate to say that PDAF + CDAF has no cost. You do lose image pixels to AF points and there can be issues AF strips showing up in photos and video.
Enabling AFF, AFC, or AF Tracking automatically lowers the burst rate. As I mentioned before, the best it can seem to do with AF is about 5.5 fps -- not the claimed 7.5 fps.
I feel better now. I noticed this and thought I was doing something wrong. Thanks for this.
Kim Letkeman
Forum Pro
Like spling? lolYes, you more than others. I am glad we are able to tech you simple things.we all were learning at some time.
Most people understand that, but I guess not all.
I must have missed something from the past ... you attacked Sergey's first comment, setting up the ruin of the thread. What's that about?
Kim Letkeman
Forum Pro
Is this one of those challenges that boils down to "let's compare wallets?"
Wrong. Single PDAF lacks of precision. In one step it can define probable location of t the subject. When it works in team with CDAF, final result is far more accurate.
DFD is the inseparable part of CDAF, just improved algorithm based on the knowledge of specifics of the lens. In fact, it can be as fast or faster than PDAF + CDAF.
Jon555
Veteran Member
I have a couple of questions, if you get a sec.
Firstly which firmware version do you have?
Secondly what's the preview AF speed like? By that I mean when you are just pointing it at things and not forcing AF. So aim at the wall then at the floor by your feet (assuming the lens focuses that close) and how fast does the floor come into focus? The ones I tried yesterday were maybe 5x slower than my GH3.
Firstly which firmware version do you have?
Secondly what's the preview AF speed like? By that I mean when you are just pointing it at things and not forcing AF. So aim at the wall then at the floor by your feet (assuming the lens focuses that close) and how fast does the floor come into focus? The ones I tried yesterday were maybe 5x slower than my GH3.
Ontario Gone
Senior Member
- Messages
- 4,183
- Solutions
- 1
- Reaction score
- 1,545
Agreed. I was very excited about DFD when i first heard because the logic seems so well thought out. Panasonic really loop holed the issues behind on sensor AF. For one, the same advantages of CDAF over PDAF will continue, such as no reduced noise performance from PDAF pixels taking up space. They won't ever have to deal with on sensor PDAF, it's simply the same CD with more advanced algorithms. Im sure there will be haters claiming this and that, but it lacks hardware requirements and thats good. Once the algorithms are created, they are there to use and shouldn't need much adjusting over time since the lenses are not changing. It really is an ingenious design.
This is also the first iteration, i can only imagine the 2nd, or 3rd generation. Or 10th. Whats also interesting is all the arguing about mirrorless vs DSLR. The wholy grail of the DSLR is AF tracking, but that edge is slipping away on multiple fronts. First the EM1, then the A6k, now GH4. Interesting times. I do think this will be faster than the combo you mention Grins, and i think it will be cheaper due to the lack of hardware needed on sensor. I wonder if anybody will try to copy Panasonic here, or if copyrights will encompass enough to prevent that.
When I was using the 100-300mm lens the cameras was extremely slow. Maybe I need new firmware? When I switched to the 35-100, it was like a machine gun! It shocked me because I wasn't expecting it. At least now I know why it seemed slow before.
Which lens were you using? If I am at a completely blank wall it takes a bit to focus or won't focus at all. If there is any texture, it focuses very fast.
Sergey_Green
Forum Pro
But it is still f/5.6 equivalent, either way you twist it. And I figure the motion was likely this comparable, am I correct?
So he was just shaking his head then? I see.
On digital photography review forum, why should I care what else you do with your camera?
--
- sergey
Jon555
Veteran Member
12-35 and 42.5, plus the floor had wooden boards so lots of stuff to focus on.
I would think a wooden floor would be easy to focus on.
This was pretty instantaneous. It was with he 35-100mm lens.
micksh6
Senior Member
Canon M was a first try, new mount, new lenses. The lenses have to be optimized too. Canon doesn't seem to care about EOS-M, so why should we? Canon 70D has better PDAF on sensor than M.
I don't have any hard evidence re: algorithms, but there is a lot of small pieces of information here and there, so I think it's valid to say that algorithms play a major role in AF.
You can find many examples of camera firmware updates for different manufacturers that improve AF, that's algorithms.
The whole thing with PDAF is that PDAF sensors give similar output to anyone who uses them.
Yet, for years Nikon cameras had better AF than Pentax, for example. Physical differences in PDAF sensors matter, but it's easy to reverse-engineer physical parts. It's the algorithms that are hard to copy.
No, there are adjustments in process, see this article, it's very interesting
http://www.dpreview.com/articles/5402438893/busted-the-myth-of-open-loop-phase-detection-autofocus
AF speed/accuracy is about how good the first estimations are. In the end, all AF systems will require final adjustments, but if the first focus position estimation is way off it will require more time to fine-tune in order to achieve good accuracy. Below I'll show why initial DFD estimation can't be as accurate as PDAF.
Theoretically yes, but in practice the benefits are questionable.
Sure, everything is a speculation here, but I'm trying to make speculations reasonable.
DFD is based on analyzing background blur, right? Can you estimate distance to background based on few shots with different focus distance? No? Why do you think small chip in camera will be able to do than in milliseconds?
PDAF sensors on the other hand, have been designed and optimized to provide phase (or angle) difference, which after some math and lens motor calibration can be translated to focus position, a single number.
I was talking about continuous AF. This was the topic of the whole thread IMO - how GH4 C-AF is good for capturing soccer game.
Predicting focus position in the next 100ms requires real-time analysis of data coming from different sensors. That's where algorithms are important.
It's easy to see why good DFD implementation will be way more difficult than PDAF+CDAF. The whole DFD concept is based on analyzing differences in blur of out-of-focus areas. It's a lot more data than few numbers that come from PDAF sensors. We are talking about area data vs single numbers, two orders of magnitude more data.
It's also easy to see why DFD data can be misleading. The blur character can be very different depending on background type. Ever heard of that "busy vs smooth bokeh" thing?
It can be difficult for human to estimate background distance depending on how detailed/contrasty the background is, even with the same lens. Surely it will be harder for machine, especially if it's a small underpowered chip and you need results in milliseconds.
As for quality loss with PDAF on sensor - It is there, but only theoretically. In practice it's negligible. I dare anyone to find cases where GH4 in the same conditions shows better image quality vs E-M1 that uses the same sensor.
Last edited:
micksh6
Senior Member
100-300mm has slow aperture motor. It can't adjust aperture fast enough for high speed shooting.
In 2012, when Panasonic cameras were unable to shoot faster than 5fps, this lens didn't allow to use full 9fps speed on Olympus E-M5, unless used wide open.
Shoot wide open with this lens for full speed. But, for better IQ you may need to stop down. 100-300mm might be slow focusing too, I can't be sure. It's just most of the good long telephoto action shots I saw were taken with Oly 75-300mm II.
You don't have to use Oly 75-300mm II wide open for full speed, BTW. But, you won't have IS, DFD or PDAF. Panasonic cameras have their compromises.
PDAF pixels strips
It would not say "theoretical" since the PDAF pixels occasionally can been seen in images.
I would say 99.99% of the time it is a non-issue.
I am holding off judgment on the PDAF vs. DFD question. I can't wait to see the 2nd and 3rd generations.
I haven't seen anyone answer my question about PDAF turning off in lower light. It my understanding that on -sensor PDAF does not work in low light. I hope someone can clear this up.
Ontario Gone
Senior Member
- Messages
- 4,183
- Solutions
- 1
- Reaction score
- 1,545
It seems there are a lot of people hoping the DFD in the GH4 doesn't do well. Im not sure why else there would be so much negativity for an AF system with very little review so far.
Keyboard shortcuts
- f
- Forum