Contrast detect autofocus - WOW

K

Kanine

Messages
183
Edit My Images
No
A reply I recently gave which I think worthy of a new thread and discussion:


I never used to use live view - until a few days ago I discovered possibly the most significant discovery I've made since using the AF-ON button...

Contrast detect/live view autofocus is THE MOST accurate way to focus compared to phase/viewfinder autofocussing.

I am not a pixel peeper, but I like my lenses to be as sharp as they can be. I don't care if corners are slightly soft or anything like that, but why not have sharpness if it's available?

Anyway...I thought I might have a ever so slightly soft 85 1.4, and wasn't expecting it to be sharp wide open really...after using AF fine tune for a while it was spot on in the centre, but on the far extreme edge focus points, it would be a little fuzzy. Fair enough I thought, edge sharpness might not be great.

WRONG. With contrast detect in live view, this lens is RAZOR sharp all over the frame, anywhere I select. It turns out that whilst the center cross AF point was super accurate, the other 36 focus points on the D300s that aren't cross points aren't anywhere as near as good, even in good light. I was so amazed that I started playing with all my other lenses that I thought were a little soft wide open - 50 1.4, 18-50 2.8 ....these lenses were now razor sharp whenever I focussed, EVERY time!

Pity the contrast focus is SO slow! But if it means nailing focussing wide open with 1.4 primes every time, I might just live with slow focus.

See how well it works for you...its made my prime experience even better!
 
Well, using all AF points and a fast prime wide open doesn't normally lead to best results, unless you get lucky. Please correct me if any camera can do it properly, - how does it know where exactly The subject is? Centre point only is a more trusty way.
But overall I agree, liveview focusing (I have to use manual focus and horrible screen on 40D) does give *excellent* results, even at 85mm f/1.8 shooting landscape shots (only for testing, but still it works!). The method is not very practical apart from use on tripod for landscapes.
 
Kanine said:
A reply I recently gave which I think worthy of a new thread and discussion:


I never used to use live view - until a few days ago I discovered possibly the most significant discovery I've made since using the AF-ON button...

Contrast detect/live view autofocus is THE MOST accurate way to focus compared to phase/viewfinder autofocussing.

I am not a pixel peeper, but I like my lenses to be as sharp as they can be. I don't care if corners are slightly soft or anything like that, but why not have sharpness if it's available?

Anyway...I thought I might have a ever so slightly soft 85 1.4, and wasn't expecting it to be sharp wide open really...after using AF fine tune for a while it was spot on in the centre, but on the far extreme edge focus points, it would be a little fuzzy. Fair enough I thought, edge sharpness might not be great.

WRONG. With contrast detect in live view, this lens is RAZOR sharp all over the frame, anywhere I select. It turns out that whilst the center cross AF point was super accurate, the other 36 focus points on the D300s that aren't cross points aren't anywhere as near as good, even in good light. I was so amazed that I started playing with all my other lenses that I thought were a little soft wide open - 50 1.4, 18-50 2.8 ....these lenses were now razor sharp whenever I focussed, EVERY time!

Pity the contrast focus is SO slow! But if it means nailing focussing wide open with 1.4 primes every time, I might just live with slow focus.

See how well it works for you...its made my prime experience even better!
Click to expand...

I noticed the exact same thing just a few days ago my friend. I noticed that nearly all the photos I produced with my 450D had to be sharpened. Not that they weren't acceptable beforehand, but for them to tack sharp they *needed* postprocessed. Now I find with contrast detect AF the amount of photos I need to post process for sharpness is vastly reduced. I guess the tradeoff of softer images with quick-AF is that it is indeed that - quick. Both have their uses, but where you've got the time, live view and contrast detect is a winner. Nice one.
 
daugirdas said:
Well, using all AF points and a fast prime wide open doesn't normally lead to best results, unless you get lucky. Please correct me if any camera can do it properly, - how does it know where exactly The subject is? Centre point only is a more trusty way.
But overall I agree, liveview focusing (I have to use manual focus and horrible screen on 40D) does give *excellent* results, even at 85mm f/1.8 shooting landscape shots (only for testing, but still it works!). The method is not very practical apart from use on tripod for landscapes.
Click to expand...

You would think that the AF on the D300 - which is essentially almost the same as the D700/D3 - 15 cross points and 36 single points - would be pretty accurate, especially on the D3! And you'd think with the £800 prime I was using, you'd be hoping for something a bit special....it seems like contrast detect is the only way to achieve this!

For contrast detect AF, you still have a little square which you can position as your focus point.

I'd also like to say it is reasonably practical, you just lose the feel good feeling of holding a camera to your eye...which is a big downer for me too, unfortunately!
 
Kanine said:
You would think that the AF on the D300 - which is essentially almost the same as the D700/D3 - 15 cross points and 36 single points - would be pretty accurate, especially on the D3! And you'd think with the £800 prime I was using, you'd be hoping for something a bit special....it seems like contrast detect is the only way to achieve this!
Click to expand...

OK, at f/1.4-1.8 the DOF is paper-thin, even at longer focusing distances. All these focusing points will be effectively pointing at different objects at different distances (eg. trees far away and sky at the proper infinity; nose, eyes, ears, neck, etc). How does the camera choose which one you need? Well it takes some average value. Only YOU can make an informed choice. Also there are reasons why focusing points towards the lens corners are less accurate - the lens focal plane may not be exactly flat, or there may be a very slight centering defect.
 
I think this is why newer cameras allow you to tweak focussing on individual lenses.
I've never really seen the need until recently I tested a lens which was exactly as you describe - soft edges unless I focussed manually using Live View. If that was my lens I'm sure a focus tweak could mean the whole frame would be sharp under auto-focus too.
 
daugirdas said:
OK, at f/1.4-1.8 the DOF is paper-thin, even at longer focusing distances. All these focusing points will be effectively pointing at different objects at different distances (eg. trees far away and sky at the proper infinity; nose, eyes, ears, neck, etc). How does the camera choose which one you need? Well it takes some average value.
Click to expand...

It achieves the most contrast in the box/point selected, the way it always works. If there's a large part of foreground and minimal background, it will pick the foreground. And vice versa. You wouldn't select the bg if you wanted the foreground in focus would you?!

daugirdas said:
Also there are reasons why focusing points towards the lens corners are less accurate - the lens focal plane may not be exactly flat, or there may be a very slight centering defect.
Click to expand...

Irrelevant surely, as the camera is doing the autofocussing. If it can achieve it one way (contrast detect) then the lens does not have a focal plane or centering problem.
 
DuncanDisorderly said:
I think this is why newer cameras allow you to tweak focussing on individual lenses.
I've never really seen the need until recently I tested a lens which was exactly as you describe - soft edges unless I focussed manually using Live View. If that was my lens I'm sure a focus tweak could mean the whole frame would be sharp under auto-focus too.
Click to expand...

I thought this was the case too, but it seems that whilst a value of tweaking + 15 for centre would be perfect, it would need only +5 for the edges...something not yet possible to program in. It's either all or nothing unfortunately. Think there's an inherent inaccuracy in single directional phase detect sensors that only cross sensors can improve on.
 
Kanine said:
I thought this was the case too, but it seems that whilst a value of tweaking + 15 for centre would be perfect, it would need only +5 for the edges...something not yet possible to program in. It's either all or nothing unfortunately. Think there's an inherent inaccuracy in single directional phase detect sensors that only cross sensors can improve on.
Click to expand...

Just give up those unnecessary AF points for anything non-sports related. Use centre point, recompose and shoot. Problem solved. I wouldn't trust using all AF points for some subjects even with D3S or 1DIV.
Also, the difference between foreground and background can be very seamless unlike you stated. The differences can be very subtle, yet the difference it could make is very significant. At the end of a day you take photos with your d300; d300 doesn't take it for you.

edit: I am also very glad to hear the Nikons famous AF is also failing. It makes me feel better using my Canon :razz:
 
daugirdas said:
the lens focal plane may not be exactly flat, or there may be a very slight centering defect.
Click to expand...
I think this is spot on - I still get a headache trying to focus my TS-E because it has been designed NOT to have a flat plane of focus.

Lenses that have soft edges and can be sharpened up by tweaking the focus are almost certainly due to a non-flat focus plane. Focussing on the mid point means just that - the middle of the DoF is optimised for the center of the lens. By tweaking the focus you are aiming to slightly de-focus the center so that that the DoF covers the edge and center, not just centered around the center.

I've no evidence for this, it's just a hunch based on what I've observed - feel free to shoot me down in flames :D
 
daugirdas said:
edit: I am also very glad to hear the Nikons famous AF is also failing. It makes me feel better using my Canon :razz:
Click to expand...

Haha I agree with that. Gives us ammunition when the Nikonites come at us :cuckoo:
 
DuncanDisorderly said:
I think this is spot on - I still get a headache trying to focus my TS-E because it has been designed NOT to have a flat plane of focus.

Lenses that have soft edges and can be sharpened up by tweaking the focus are almost certainly due to a non-flat focus plane. Focussing on the mid point means just that - the middle of the DoF is optimised for the center of the lens. By tweaking the focus you are aiming to slightly de-focus the center so that that the DoF covers the edge and center, not just centered around the center.

I've no evidence for this, it's just a hunch based on what I've observed - feel free to shoot me down in flames :D
Click to expand...

Not entirely sure what you mean. Understand if it's not sharp in the corners if you focus in the CENTRE.

But if you focus IN the corners/edges and it's not sharp there, (but it is possible for it to be sharp via manual focus), how is it the lenses fault? It's the camera not focussing properly (which contrast detect fixes and where phase detect fails in my experience miserably..)
 
daugirdas said:
. At the end of a day you take photos with your d300; d300 doesn't take it for you.
Click to expand...

Sorry, I don't understand your point, unless you're advocating only manual focus?! I choose what to focus on (or something the same distance to focus on) to achieve focus.
 
Kanine said:
Probably THE worst technique you could use to achieve focus, especially with razor sharp DoF. :p

http://visual-vacations.com/Photography/focus-recompose_sucks.htm

If you've tried it with a 1.4 prime, you'll have learned it's not possible to achieve sharp focus with recomposing off centre!
Click to expand...

I have 1.8 prime, and it works 100%. I also use 100mm f/2.8 macro, and have used it at 1:1 wide open and also worked just fine. Practise your technique maybe?
 
daugirdas said:
I have 1.8 prime, and it works 100%. I also use 100mm f/2.8 macro, and have used it at 1:1 wide open and also worked just fine. Practise your technique maybe?
Click to expand...

:p

It's not technique, if you read the link, you'll realise it's physics.
 
Kanine said:
Sorry, I don't understand your point, unless you're advocating only manual focus?! I choose what to focus on (or something the same distance to focus on) to achieve focus.
Click to expand...

No. You point-press AF-On-and shoot. Sorry for sarcasm.
 
daugirdas said:
I have 1.8 prime, and it works 100%. I also use 100mm f/2.8 macro, and have used it at 1:1 wide open and also worked just fine. Practise your technique maybe?
Click to expand...

technique aside putting an AF point over an eye is a sure fire way to abalutely nail focus, I use any point that suits the composition. The exception being in low light the centre point is more accurate on certain bodies

I use live view manual focussing for anything tripod mounted as its immense, but dislike the woblyness you get with not using the viewfinder (again in massively poor light (like tonight f1.4 1/30 iso 256,000) I used live view and accepted the reduced hit rate as I was having trouble focusing manually and the camera couldn't AF with any point
 
Kanine said:
:p

It's not technique, if you read the link, you'll realise it's physics.
Click to expand...

I've read the link and it is rubbish; it wouldn't stand to a peer review. If the lens focal point stays constant (the technique), then the focused subject will remain at a constant focal distance no matter how much the camera is rotated, unless the lens focal plane is spherical. It works a treat with all longer lenses (50mm+ FF equiv), I could see why a wideangle could be a problem.
 
daugirdas said:
I've read the link and it is rubbish; it wouldn't stand to a peer review. If the lens focal point stays constant (the technique), then the focused subject will remain at a constant focal distance no matter how much the camera is rotated, unless the lens focal plane is spherical.
Click to expand...

I'm not sure how you could have read the link and come to that conclusion.

The whole point is that rotating causes changes in distance as the focal plane is FLAT.

Imagine a circle with a centre point. Now pick ANY point on the outside of that circle and draw a line from the centre to that point. Now draw a tangent to that line. Pick another point on the circle, close to the first point. Repeat the process.

See how the two tangents CANNOT coincide? Hence why the subject STILL cannot be in focus if you recompose. Again, simple physics.

daugirdas said:
It works a treat with all longer lenses (50mm+ FF equiv), I could see why a wideangle could be a problem.
Click to expand...

If anything it is WORSE with telephoto lenses as the DoF is even shallower at close ranges (i.e. portraits), especially fast teles. Wideangles should be able to get away with any recomposition errors probably, due to the inherent deeper DoF.
 
david1701 said:
putting an AF point over an eye is a sure fire way to abalutely nail focus, I use any point that suits the composition.
Click to expand...

It sounds very good, but with 11 points around frame centre it could be very limiting. Some points may be significantly less accurate which I have noticed. Lastly, switching between different points can be a very daunting task.
There is also some indication in the viewfinder if the things go wrong. It wouldn't be very wise to ignore the obvious.
 
I realised this also some time ago, and it helped me no end, because if my Sigma 10-20mm has a weak point, it's that I always get a bad focus despite the beep. I generally use this lens for though out shots, wedding groups, wide angle building stuff, so t doesn't lose me anything with the slow contrast detect AF. I certainly hope this does become the future of AF. I fnothing else, it may narrow down the distance between mid-range and top-end lenses in terms of purely AF operation.
 
Kanine said:
I'm not sure how you could have read the link and come to that conclusion.

The whole point is that rotating causes changes in distance as the focal plane is FLAT.

Imagine a circle with a centre point. Now pick ANY point on the outside of that circle and draw a line from the centre to that point. Now draw a tangent to that line. Pick another point on the circle, close to the first point. Repeat the process.

See how the two tangents CANNOT coincide? Hence why the subject STILL cannot be in focus if you recompose. Again, simple physics.
Click to expand...

You are right, but you try to prove the wrong thing. Imagine a circle with 2 points one in a centre and one on the edge. If you rotate the the circle, the distance between the points, or any other point on the perimeter doesn't change. Other distances WILL change, but they should be not critical for the composition. Otherwise a different angle of f-stop may be required.
That is the whole point of focusing and recomposing. Once the focus is locked, the focal point (the sensor) is kept fixed, and the camera is rotated around an imaginary axis going through the point. If there is any movement offsetting any of the 2 points, there will be a focus error, that can be fixed by manual focus tweaking.



If anything it is WORSE with telephoto lenses as the DoF is even shallower at close ranges (i.e. portraits), especially fast teles. Wideangles should be able to get away with any recomposition errors probably, due to the inherent deeper DoF.
Click to expand...

WRONG. I bet you didn't really try it. DOF is shallower in general, but the subjects tend to be further away. DOF increases with focal distance, and at infinity we get plenty of DOF (or at macro distances extremely shallow DOF). It actually compensates the user error pretty nicely. The corrections to recompose also much smaller as compared with wideangles.
 
daugirdas said:
It sounds very good, but with 11 points around frame centre it could be very limiting. Some points may be significantly less accurate which I have noticed. Lastly, switching between different points can be a very daunting task.
There is also some indication in the viewfinder if the things go wrong. It wouldn't be very wise to ignore the obvious.
Click to expand...

Switching between AF points is as difficult as...moving your thumb?
Theyre also pretty inaccurate as they only measure contrast along one or two thin lines...whereas it certainly seems that contrast detect uses the whole box to determine the best focus (I think, anyway) which is a lot more accurate.

I'd NEVER rely on the viewfinder for focus accuracy, firstly as I'd much rather trust a 100% zoom on playback rather than my eyesight, and secondly as the majority (if not all) of DSLR bodies only show DoF limited to 2.5 regardless if the lens is faster. Using anything faster than 2.5 means you can't actually see the result without taking the exposure, thus you won't be able to see if your subject is accurately in the DoF. This is also why the bokeh is much, much better on the final pic than what you can see through the VF. (unless you replace the VF with a specially designed screen)

Anyway, I'm not going to persuade you anymore, it seems like you've made up your mind. I'm not here to force anyone to do anything!

For everyone else though, thought it might be a useful snippet of information to maximise your image quality - give it a try, it might just work for you ;)
 
Phew! After reading through this thread I'm glad I don't have any focusing issues like that! :)
 
daugirdas said:
You are right, but you try to prove the wrong thing. Imagine a circle with 2 points one in a centre and one on the edge. If you rotate the the circle, the distance between the points, or any other point on the perimeter doesn't change. Other distances WILL change, but they should be not critical for the composition. Otherwise a different angle of f-stop may be required.
That is the whole point of focusing and recomposing. Once the focus is locked, the focal point (the sensor) is kept fixed, and the camera is rotated around an imaginary axis going through the point. If there is any movement offsetting any of the 2 points, there will be a focus error, that can be fixed by manual focus tweaking.
.
Click to expand...

If something is in the centre of the frame (lets say 10cm away) then the edges of the frame will be now focussing on a distance slightly further (lets say 11cm) as the plane of focus lies at a tangent to the focus distance. Recompose after focussing, with subject now on edge of frame. What happens? Subject is still 10cm away but subject to a focus field of 11cm away. Unsharp subject.....



daugirdas said:
WRONG. I bet you didn't really try it. DOF is shallower in general, but the subjects tend to be further away. DOF increases with focal distance, and at infinity we get plenty of DOF (or at macro distances extremely shallow DOF). It actually compensates the user error pretty nicely. The corrections to recompose also much smaller as compared with wideangles.
Click to expand...

Hence why I said at CLOSE distances with FAST primes. Obviously anything far away won't be affected nearly as much.
 
This thread has lost focus! :D (Pun intended)

My point is:

Contrast detect AF can help you to maximise your image quality and sharpness - give it a try, it might just work for you.
 
daugirdas said:
It sounds very good, but with 11 points around frame centre it could be very limiting. Some points may be significantly less accurate which I have noticed. Lastly, switching between different points can be a very daunting task.
There is also some indication in the viewfinder if the things go wrong. It wouldn't be very wise to ignore the obvious.
Click to expand...

thats what cropping is for, I'd rather crop a perfect image to tweak the comp than have a beautifully composed slightly oof image. The points are laid in the most useful conditions for standard things aswell which means they are usually spot on

aye some are less accurate, thats why if i think an outer point will be inaccurate I use the centre AF point and crop to sort the composition.

its basically instant to switch between them using the joystick on any xxd or above body (nb I have only tried this on 30d and later and not on 1d's so really its 30/40/50/5dII I can guarantee its really really easy on)

If handheld using live view and contrast AF introduces too many errors in camera and subject movement, negating the advantages in accuracy. On a tripod it must be great but on a tripod its just as easy and a lot quicker to zoom in 5x/10x and manually focus to perfection
 
Kanine said:
This thread has lost focus! :D (Pun intended)

My point is:

Contrast detect AF can help you to maximise your image quality and sharpness - give it a try, it might just work for you.
Click to expand...

And I think it is a very good point, although trying to persuade some folks that their DSLR (full of old work-around technology) is anything less than perfect, can be difficult :bang:

Contrast detect AF is fundamentally better than phase-detect, if only because it reads off the sensor and not an approximated equivalent position. It can do all sorts of things, such as accurate focus anywhere in the frame as you've mentioned.

Also, once it has locked on to an area of the subject, it can very easily track that exact spot if it moves, all over the frame. That is a revolutionary feature for sports photographers, biffers, for anyone that shoots moving subjects.

The only small problem ( :eek: ) is that it's slow and doesn't work well in low light, but that will soon get fixed :)
 
daugirdas said:
I've read the link and it is rubbish; it wouldn't stand to a peer review. If the lens focal point stays constant (the technique), then the focused subject will remain at a constant focal distance no matter how much the camera is rotated, unless the lens focal plane is spherical. It works a treat with all longer lenses (50mm+ FF equiv), I could see why a wideangle could be a problem.
Click to expand...

Sorry to continue the off topic discussion, but you're wrong here IMHO - focus+recompose does introduce focus errors, which may or may not be significant depending on the situation.

By moving the camera up and down (when recomposing) you are describing an arc with the point of focus. However, the subject is usually flat. So the focus point moves further in front of the subject depending how much you recompose, introducing focusing errors.

Your mistake above is to say that the object remains a constant distance away from the lens as you move the camera up and down. It doesn't. Any deviation from the horizontal by the camera will effectively make the subject to lens distance greater - the source of the focusing errors.
 
daugirdas said:
I've read the link and it is rubbish; it wouldn't stand to a peer review. If the lens focal point stays constant (the technique), then the focused subject will remain at a constant focal distance no matter how much the camera is rotated, unless the lens focal plane is spherical. It works a treat with all longer lenses (50mm+ FF equiv), I could see why a wideangle could be a problem.
Click to expand...

You have contradicted yourself.

But it's not a question of focal length directly, it's a question of distance relative to depth of field.

If your DoF is measured in mm, and your focus-recompose technique moves the camera/lens-to-subject distance as few mms, as it must do, then you are going to get a big focus error.

It's very common with close distance portraits at f/1.8-1.4, when you focus-recompose on the eyes which are off centre. When you recompose, the point of sharpest focus has moved to the end of the nose.
 
martsharm said:
Sorry to continue the off topic discussion, but you're wrong here IMHO - focus+recompose does introduce focus errors, which may or may not be significant depending on the situation.

By moving the camera up and down (when recomposing) you are describing an arc with the point of focus. However, the subject is usually flat. So the focus point moves further in front of the subject depending how much you recompose, introducing focusing errors.

Your mistake above is to say that the object remains a constant distance away from the lens as you move the camera up and down. It doesn't. Any deviation from the horizontal by the camera will effectively make the subject to lens distance greater - the source of the focusing errors.
Click to expand...

I agree.

If you watch the B&H Canon videos the Canon man said exactly the same thing, which is why they consider most focusing 'issues' are user generated.
People have a tendency to focus on a subject, lock focus then recompose by rotating rather than moving along the camera sensor plane thereby introducing focus errors, especially where small apertures are used.
 
You must log in or register to reply here.
Back
Top