Focus recompose at f/0.95. (MERGED)

[jaapv]

Advertisement (gone after registration)

Hello Lord Fluff,

 

Thank you for your Post #51.

 

One of the advantages to having been married to a number of lawyers is there are times when what is happening is clear.

 

One of my wives, I think it might have been one who was a lawyer, maybe not, once said to me that the reason you never go to talk to the other person's lawyer when you are trying to settle a disagreement is you are going to educate, clarify and reasonably resolve the situation & the other person's lawyer has been hired to win.

 

As I said previously repeating something which is inaccurate does not make it more accurate.

 

In your Post #51 you refer to a statement I made earlier which I clarified specifically in my Post # 23 which you specifically responded to in your Post # 24.

 

You can't have it 2 ways. Once you acknowledge my clarification, which you did in your # 24 it is not appropriate to bring it up in # 51 as if it had never been discussed.

 

You are a very smart person. It is clear there is a lot you can add to this Forum & your input is impotant. What is the purpose of creating a battlefield in an environment which was designed for learning?

 

Best Regards,

 

Michael

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[Lord Fluff]

Hello Lord Fluff,

 

You can't have it 2 ways

 

Yet oddly this is precisely what you're trying to do when you ask "who said that? it wasn't me!", when in fact it was.

 

Part of the quest for truth is to battle against inaccuracy and misinformation, which is what I felt you (and a few others perhaps) were spreading. I have no wish (or time) for personal battles, and it's a shame if this 'argument' between us should be seen us such, but it could be time for us both to hit "unsubscribe".

[Lindolfi]

There is love for the subject, no doubt.

There is knowledge about the subject, no doubt.

There is a wish to learn, no doubt.

But there is also often a strong presence of ego's and that makes these kinds of threads dissolve into something else than the subject. These situations hardly ever produce something fruitful.

 

So could we please get back to the subject at hand: "Re-framing while shooting fast lenses wide open."

 

In my mind, we have developed some new information here or at least isolated information that was already known that can be useful while actually shooting images. In summary

 

[1] Reframing after focussing on an object with a central point in the frame (such as a rangefinder patch) does move the plane of sharpest focus away from the object: it brings the plane of sharpest focus behind the original focus point.

 

[2] This error in focussing is less problematic with small aperture, but it certainly is problematic at a wide aperture

 

[3] The error in focussing is less problematic in longer lenses due to the more narrow cone of view

 

[4] To compensate, the camera should be moved away a bit from to the object. This "a bit" can not easily be quantified while making the exposure.

 

[5] The movement of a handheld camera is helpful in getting rid of the error (see this figure), but does not remove the problem completely.

[Guest #12]

OK, here is the promised drawing. Focussing on the eye of the frog with the middle of the frame (where the rangefinder patch is) and turning the camera around the front focal point ( green dot, left hand image) creates a larger focus error than when rotating the camera around the spine (green dot, right hand image)

 

 

am not so sure about this... did you try it on graph paper, or prove it geometrically?

[Guest #12]

I got e' / e = (u + x) / u, where u is the subject distance, x is the distance between two pivots (one for the lens principal point and one for the human body), e' is the error from pivoting the body, and e is the error for pivoting the lens.

 

So (here at least) the error from rotating the equipment is always less than rotating the body behind it. Since u is large compared to x, it doesn't make that much difference.

[Bo_Lorentzen]

Sometimes, I dread to think how a written instruction for riding a bicycle would be created in a online forum.

 

Was this not originally a simple question about how to work with re-framing fast lenses.?

 

We all do it all the time, and strangely the rocking variable seems to become second nature after a little while. the error is always going to be in there somewhere as my rocking is not calibrated to mm accuracy, but seems to be good enough to get sharp eyes wide open. (the lens not the eyes.)

 

As for the accuracy, the M is a 1954 camera design, with some flaws, so is my bow and arrow, one kind of gotta get a feel for it, yet it works surprisingly well.

 

Far less important, somebody today actually told me that healthcare should not cover contraceptives because the punitive value of pregnancy prevents woman from running around having sex all over the place... I'm currently contemplating the possibility that I might accidentally have been teleported to year 1611...! unfortunately the fact Im writing this on a computer suggest Im still in 2011 America :-)

 

Bo - preoccupied with taking somewhat flawed images.

[Guest #12]

Advertisement (gone after registration)

...

We all do it all the time, and strangely the rocking variable seems to become second nature after a little while. the error is always going to be in there somewhere as my rocking is not calibrated to mm accuracy, but seems to be good enough to get sharp eyes wide open. (the lens not the eyes.)

...

 

you rock one direction, Jaap rocks the other--your pictures are both in focus because

unless you are using a short focal length, or reframing all the way to the edge, the error is not that big.

[Lindolfi]

am not so sure about this... did you try it on graph paper, or prove it geometrically?

 

Well, #12, I've calculated it using vector algebra and put it into software and produced graphs.

 

I got e' / e = (u + x) / u, where u is the subject distance, x is the distance between two pivots (one for the lens principal point and one for the human body), e' is the error from pivoting the body, and e is the error for pivoting the lens.

 

So (here at least) the error from rotating the equipment is always less than rotating the body behind it. Since u is large compared to x, it doesn't make that much difference.

 

You formula is a simplification since it does not contain focal length, which is of importance. Your formula produces roughly the inverse what I get (depending on focal length). The figure I posted illustrates the effect clearly. Could you please check the derivation of your formula and check my figure if we mean the same by "rotating the body"?

If anyone is interested I can give graphs and offer MatLab code.

 

Another point to consider: if the camera would move parallel to the plane of the sensor (without rotation) the focus-recompose error would be zero, since the plane of sharpest focus would remain in place. Moving a camera without rotation is the same as rotating the camera around a point very far (infinitely far) behind you. This is another way to show that if you move the point of rotation further away from the object (like in your spine), the focus error becomes smaller.

[Lindolfi]

Another point in the focus-recompose issue is that of DOF. Clearly with a larger DOF, the problem is less when defocussing occurs. At an equal distance, the DOF is inversely related to the square of the focal length. I did calculations for an object at 1 meter, varying focal length between 12 and 135 mm and it appears that a maximum in unsharpness occurs at a focal length of 40 mm. when focussing with the center of the frame and reframing the object to the edge of the image. (all with a 24x36mm sensor or film)

 

But all focal lengths suffer, the long lenses because of a shallow depth of field and the wide lenses because of the turning angle of the plane of sharpest focus.

 

At larger object distances, the most problematic focal length shifts to a longer focal length. For instance at 2 meter, the worst focal length is 50 mm. And at 5 meter it is 75 mm focal length.

 

When rotation of the camera around the spine is included, these worst focal lengths still apply, but the error is less.

[Guest #12]

...

You formula is a simplification since it does not contain focal length, which is of importance. Your formula produces roughly the inverse what I get (depending on focal length). The figure I posted illustrates the effect clearly. Could you please check the derivation of your formula and check my figure if we mean the same by "rotating the body"?

If anyone is interested I can give graphs and offer MatLab code.

...

 

Thanks, you're right, not the same problem. What expression did you get for the one error (from rotating the equipment), and the other error (from rotating the body)?

[adli]

The only cameras designed for focus-and-recompose are those with multiple AF points;

 

To be a bit picky, if you have multiple AF points, you don't focus before you compose, as the point of multiple AF points is to focus with the subject outside the center of the frame? Thus you compose and focus, but focus on a point off center?

[Lindolfi]

What expression did you get for the one error (from rotating the equipment), and the other error (from rotating the body)?

 

In this pdf: click you can find the expression using your variable names. If you make x zero, you get the error from rotating the camera. From the expression you can see that with an increasing x (distance between camera and pivot point), the error becomes smaller. When x grows to infinity (camera not rotated but only translated), the error becomes zero.

[wildlightphoto]

To be a bit picky, if you have multiple AF points, you don't focus before you compose, as the point of multiple AF points is to focus with the subject outside the center of the frame? Thus you compose and focus, but focus on a point off center?

 

AF focus points don't cover the entire picture area. If your composition conforms to the expected AF points and the chosen focus point works adequately well in the ambient light and with the maximum lens aperture, no worries. OTOH if your composition doesn't fit the expected patterns you might be in for a frustrating experience.

[bocaburger]

AF focus points don't cover the entire picture area. If your composition conforms to the expected AF points and the chosen focus point works adequately well in the ambient light and with the maximum lens aperture, no worries. OTOH if your composition doesn't fit the expected patterns you might be in for a frustrating experience.

 

While ago I had a Canon 1D (the original one) which used the 1V's same 45-point AF array except on a 1.3x-crop, which meant it covered most of the picture area save for maybe 25% around the periphery. It was about the closest thing to full-coverage AF there is. The problem was, Canon forgot to include the cable that tethers the camera to my brain, and so I had to rely on the camera's decision as to what the subject should be, and we disagreed often (Manually picking from 45 sensors, or even groups thereof, was uselessly slow for me). Eventually I opted to switch off all the sensors except the middle one, and use the custom function that assigns AF to a back button instead of the shutter release. That way I could ballpark the focus with AF and then fine-tune manually on the screen (many of Canon's AF lenses allow full-time manual focus w/out switching buttons). Most of the time I found myself just manually focusing.

 

My conclusion is that AF works better than manual for me with subjects that are moving both quickly and randomly, but only because I'm willing to sacrifice some compositional accuracy in favor of a sharp subject, and (if necessary) do some cropping in the computer for better composition.

[Guest #12]

In this pdf: click you can find the expression using your variable names. If you make x zero, you get the error from rotating the camera. From the expression you can see that with an increasing x (distance between camera and pivot point), the error becomes smaller. When x grows to infinity (camera not rotated but only translated), the error becomes zero.

 

Thanks.

 

A couple of us have posted the same formula before (x=0 case), so that is encouraging.

 

I'm not sure the improvement (pivot with the body) adds much. I used 0.15m for the distance between pivots, so for M lenses I think the new error is at least 90% of the old error, usually more. (135 format.) Also, unless you're using a tripod, you're probably pivoting your body anyway. In the diagram, it looks like the subject is much closer than closest focus just to show the improvement.

 

Thank you for some interesting posts and the extra info.

[Lindolfi]

Well I'm glad you agree about the improvement of defocus by a pivot point behind the camera and we agree on the expression. 0.15 m. is not realistic for a grown person (established by filming the movement of focus-recompose from above the person), it is more like 0.3 m. depending on the posture of the photographer. But that is a detail.

 

The nice thing about the expression is that when you include DOF depending on focal length (at a certain object distance), you start to understand why certain combinations produce problems with focus-recompose in practice.

 

My drawing was indeed meant to show the principle. From the formula you can derive all situations numerically, but it is better to put time in practicing adjusting the camera movement. With digital you get a nice short learning cycle. :-)

[Guest #12]

...

 

Even with that change, the improvement is about 85%, usually more, of the old error. Why not just tell people to slide laterally...or do they tend to make things worse with a complicated movement?

 

(I would still like to look at the code, if that's no trouble.)

[Lindolfi]

When only sliding sideways the composition changes: the foreground-object-background relation shifts also. It is better to learn to combine rotation with moving slightly backward. With an object distance of 1 m. and a focal length of 50 mm, you have to move only 50 mm backward when moving the object to the edge of the frame.

 

#12, here is the zip file with the source code and produced figures: click

 

Run "plotfocusrecompose.m" in MatLab, the figures will be produced.

"focusrecompose.m", which is called by "plotfocusrecompose.m", contains the vector algebra and the formula I gave in an earlier post.

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[lct]

With a leica M, if your subject is off centre, and fairly still, then focus upon it then re-frame. Do not touch the focus and all will be just fine...

Works for me as well.

[lct]

With a leica M, if your subject is off centre, and fairly still, then focus upon it then re-frame. Do not touch the focus and all will be just fine...

Works for me as well.