Architecture. Shift lens compared to post processing test.

[Nigvo]

Today I took my 24mm TS lens to a local castle to see if there are any visual differences between using a shift lens, and correcting keystoning in post. The lens was on a tripod and in a fixed position. In one case the lens was shifted, in the other, shift was set to zero and the lens tilted to simulate a normal lens.

I looked at two cases: simple vertical shift and diagonal shift. I then made an approximate overlay of the images to get an idea of the differences.

Simple vertical shift is pretty close, with very minor differences probably due to my approximations. Diagonal shift does have some more noticeable differences.

I just wanted to do a geometrical comparison. The merits of using a shift lens compared to correcting in post are not the subject of this post.


24 TS lens shifted vertically


24mm TS no shift, lens tilted


Overlay


24mm Diagonal shift


24mm TS Lens with zero shift tilted upwards and corrected in post. 2 way correction


Overlay


24mm TS Diagonal shift


24mm TS Lens with zero shift tilted upwards and corrected in post. 2 way correcton




Overlay









“Everyone sees what you appear to be, few experience what you really are.”
- Niccolo` Machiavelli.

 

[chrisfisheye]

Hello,

Not sure people will understand your logic. I understand it more or less becauee I have more context when I had the opportunity to discuss with you in another thread.

First remark, if I can understand that vertical shift can improve an image this is very rarely a good idea to do horizontal shift. It can become very quickly unnatural.

Secondly the results can be perfectly calculated so both methods will give the exzct same result. Except that T&S are not corrected, so even if they usually have low barrel/pincushion distorsion they are not perfect. Even a bzd normal lens would have the geometry more correct than a T&S because it is corrected. I am not talking about IQ of course, T&S are usually very good, this is not the question, I am just explaining why the results may not match precisely.

 

[Nigvo]

These are the post corrected shots before and after post correction.





















--
“Everyone sees what you appear to be, few experience what you really are.”
- Niccolo` Machiavelli.

 

[Nigvo]

Hello,

Not sure people will understand your logic. I understand it more or less becauee I have more context when I had the opportunity to discuss with you in another thread.

The logic is quite simple. I wanted to compare a shot taken with a TS shifted to correct the verticals, compared to normal lens tilted and the verticals subsequently corrected in post.

First remark, if I can understand that vertical shift can improve an image this is very rarely a good idea to do horizontal shift. It can become very quickly unnatural.

Take a look at the pictures and you will see two things.

First the simple vertical shift gives images which are very similar to those corrected in post with the vertical keystone tool.

When we move onto diagonal shift, we see some differences. Some stretching starts to occur in the post "two way" corrected image-

The optical version will not be unnatural, because we are using a portion of a highly corrected lens with a very large image circle. My examples show this to be true.

The post corrected version starts to have problems with stretching on the most highly corrected side of the image.

Secondly the results can be perfectly calculated so both methods will give the exzct same result. Except that T&S are not corrected, so even if they usually have low barrel/pincushion distorsion they are not perfect. Even a bzd normal lens would have the geometry more correct than a T&S because it is corrected. I am not talking about IQ of course, T&S are usually very good, this is not the question, I am just explaining why the results may not match precisely.

TS lenses are very costly because they are better optically corrected for barrel or pincushion distortion. I used the same 24mm TS for my tests, which should bring the results much closer, than if I had used a lens with less well controlled distortion.

 

[John Crowe]

Oh, so there was a minimal amount of camera tilt used. I have been in situations where considerably more was required with a 24mm lens.

In this mild case there is not likely much distortion and loss of resolution when looking at details, but do you see evidence of that.

I am curious why the yellow building beside the tower does not appear to be perpendicular in any image, except the diagonal shifted one. Is it just my perception, here on my phone?

 

[Nigvo]

Oh, so there was a minimal amount of camera tilt used. I have been in situations where considerably more was required with a 24mm lens.

I probably need to find a subject where I need full shift.

In this mild case there is not likely much distortion and loss of resolution when looking at details, but do you see evidence of that.

I have not pixel peeped, but the resolution loss is easy to calculate at around 10% - 30% in most cases, and even ore when you leave space for cropping to the same view.

I am curious why the yellow building beside the tower does not appear to be perpendicular in any image, except the diagonal shifted one. Is it just my perception, here on my phone?

The Yellow building does indeed have walls that lean inwards. It is an old restored building.

 

[Klaus dk]

Very informative. Thank you for sharing.

 

[ggbutcher]

When we move onto diagonal shift, we see some differences. Some stretching starts to occur in the post "two way" corrected image-

I know just enough about such math to be dangerous, so take this with a grain of salt...

Image transforms that change the geometric relationships of the pixels that I know do them in two steps, row-wise, then column-wise, or vice versa. In the intermediate between the two would be a place where such variation could occur. So, it might be useful to find out what you can about how your software is doing the transform...

 

[Joe Bagadonuts]

In your example, you actually are doing very little shift, and have the ability to stand way back, both being an easy layup for either method. Try shooting that tower at full shift from way closer and then do it with software and see why software isn’t perfect. You can’t frame correctly when you don’t have plenty of relief around the subject. The picture starts to fall apart more with big software shifts.

 

[Nigvo]

In your example, you actually are doing very little shift, and have the ability to stand way back, both being an easy layup for either method. Try shooting that tower at full shift from way closer and then do it with software and see why software isn’t perfect. You can’t frame correctly when you don’t have plenty of relief around the subject. The picture starts to fall apart more with big software shifts.

I might try to do some more extreme shifts. But I was trying to examine a real life situation.

But using a well corrected lens probably played its part too, as any barrel distortion in a normal less corrected lens will be exaggerated if you correct in post. Add in a lens which has correction applied in post ( most M43 lenses for example) and image quality will fall even further.

 

[MoreorLess]

In your example, you actually are doing very little shift, and have the ability to stand way back, both being an easy layup for either method. Try shooting that tower at full shift from way closer and then do it with software and see why software isn’t perfect. You can’t frame correctly when you don’t have plenty of relief around the subject. The picture starts to fall apart more with big software shifts.

I might try to do some more extreme shifts. But I was trying to examine a real life situation.

But using a well corrected lens probably played its part too, as any barrel distortion in a normal less corrected lens will be exaggerated if you correct in post. Add in a lens which has correction applied in post ( most M43 lenses for example) and image quality will fall even further.

Really though a lot of the reason you buy a shift lens is because your shooting close to tall subjects, if thats all you plan on doing honestly I'm not sure its worth it over correcting a standard lens.

 

[chrisfisheye]

Hi,

I hope you are now convinced that perspective correction can reproduce the exact result of a T&S lens (in terms of geometry I mean).

It is true that some tools do not reproduce the exact result by default. It can correct verticals, horizontals, but the proportion of objects may be different. You have other sliders to change the aspect for instance.

I really regret it works this way but shifting does distort somewhat the geometry anyway, so the goal is not necessarily to reproduce exactly the T&S result but maybe more to get a pleasing result. I repeat again that it works anyway with verticals and the image if the church you had posted does not proove the opposite.

I certainly prefer to reproduce the T&S result, it is in my opinion the most correct one. I this case Hugin and certainly other tools like PtGui wotk perfectly. You can post other examples if you prefer but maths don't lie. So even diagonal shift can be reproduced without difficulties.

To understand why it can be computed. Put the camera at the center of a sphere, with the sensor at distance=focal length. Place the image on the sensor and project the image on the sphere (as you can see in the Hugin screenshot I have posted). Rotate the camera at will (pitch, yaw, roll) and then project again the image from the sphere on the sensor. This is a perfect transformation.

This does not mean I consider that T&S are useless, they may be a bit better but I don't see how they could really outperform the results from software corrections. There are not horrible and unnatural results on one side and fantastic results on the other side.

 

[Nigvo]

Hi,

I hope you are now convinced that perspective correction can reproduce the exact result of a T&S lens (in terms of geometry I mean).

I proved that for diagonal shift, it does not.

It is true that some tools do not reproduce the exact result by default. It can correct verticals, horizontals, but the proportion of objects may be different. You have other sliders to change the aspect for instance.

Why not just get it right in the camera with the right lens for the photo. Job done.

I really regret it works this way but shifting does distort somewhat the geometry anyway, so the goal is not necessarily to reproduce exactly the T&S result but maybe more to get a pleasing result. I repeat again that it works anyway with verticals and the image if the church you had posted does not proove the opposite.

No. TS lenses are well corrected for distortion. Shifting does not distort . You are using just a portion of a lens with a large image circle. Read any book about large format photography that use cameras with movements as a default.

So geometrically you are not distorting anything to any significant degree.

I certainly prefer to reproduce the T&S result, it is in my opinion the most correct one. I this case Hugin and certainly other tools like PtGui wotk perfectly. You can post other examples if you prefer but maths don't lie. So even diagonal shift can be reproduced without difficulties.

Most Architectural photographer disagree strongly.

I quote from James Ewing from his textbook Follow the Sun. A book aimed at professionals and students and considered a standard text.

"You might ask yourself "Do I really need an expensive tilt shift lens" Can't I just correct the perspective later in Photoshop?" The answer is yes you could correct it later, but the tilt shift lens allows you to see and feel the perspective of the images you are shooting. The final crop and ultimately the entire composition will be totally different in a shot that is corrected in post. If you cannot see the image while you are shooting you cannot control the composition and therefore you cannot effectively interpret the building. Correcting the perspective during post production causes a significant loss of sharpness and detail. The Tilt shift lens gives you accurate, sharp controlled images."

I think this says it all.

To understand why it can be computed. Put the camera at the center of a sphere, with the sensor at distance=focal length. Place the image on the sensor and project the image on the sphere (as you can see in the Hugin screenshot I have posted). Rotate the camera at will (pitch, yaw, roll) and then project again the image from the sphere on the sensor. This is a perfect transformation.

This does not mean I consider that T&S are useless, they may be a bit better but I don't see how they could really outperform the results from software corrections. There are not horrible and unnatural results on one side and fantastic results on the other side.

Well for a start I use the whole sensor area and do not have to crop away significant portions of the image or stretch the image.

 

[Dspider]

Hi Mate, whats your thoughts on this thread

Loading…

www.dpreview.com

Ds

 

[Reilly Diefenbach]

Hi,

I hope you are now convinced that perspective correction can reproduce the exact result of a T&S lens (in terms of geometry I mean).

It is true that some tools do not reproduce the exact result by default. It can correct verticals, horizontals, but the proportion of objects may be different. You have other sliders to change the aspect for instance.

I really regret it works this way but shifting does distort somewhat the geometry anyway, so the goal is not necessarily to reproduce exactly the T&S result but maybe more to get a pleasing result.

Or even a much more correct result. It's never going to be perfect, but there are a lot more choices to be made than a tilt shift lens can provide.

I repeat again that it works anyway with verticals and the image if the church you had posted does not proove the opposite.

I certainly prefer to reproduce the T&S result, it is in my opinion the most correct one. I this case Hugin and certainly other tools like PtGui wotk perfectly. You can post other examples if you prefer but maths don't lie. So even diagonal shift can be reproduced without difficulties.

To understand why it can be computed. Put the camera at the center of a sphere, with the sensor at distance=focal length. Place the image on the sensor and project the image on the sphere (as you can see in the Hugin screenshot I have posted). Rotate the camera at will (pitch, yaw, roll) and then project again the image from the sphere on the sensor. This is a perfect transformation.

This does not mean I consider that T&S are useless, they may be a bit better but I don't see how they could really outperform the results from software corrections. There are not horrible and unnatural results on one side and fantastic results on the other side.

Well, that's pretty much how I see it. Unnatural distortions from the tilt/shift, in this case the top corners racked to a point completely unlike what the viewer would experience at the venue.


Google Images



much improved with software:






Photoshop

If you can't "get it right in camera" because of the problematic opticals, there is no other recourse than fixing it in software.



The OP has put up a "test" taken far away which is no challenge at all for any lens. Not going to fly at all if you're across the street from a large building (you usually are) shooting up.

 

[chrisfisheye]

Hi,

I hope you are now convinced that perspective correction can reproduce the exact result of a T&S lens (in terms of geometry I mean).

I proved that for diagonal shift, it does not.

You prooved absolutely nothing, come on.

If maths don't lie, images often lie..

Each time I propose that you go to the scientific forum, you refuse. Why ???

You perfectly know you are not correct of course.

. Shifting does not distort .

Anybody (except you maybe) using a T&S lens knows that it creates perspective distorsion for 3D scene. How can't you see this ?

It takes 2 seconds to check, use maximum vertical + horizontal shift and look... It can work work more or less with simple architecture but as soon as there is relief like balconies) it is not good at all. Better to use it with moderation or step back when you can and not use it.

 

[chrisfisheye]

Hi,

I hope you are now convinced that perspective correction can reproduce the exact result of a T&S lens (in terms of geometry I mean).

It is true that some tools do not reproduce the exact result by default. It can correct verticals, horizontals, but the proportion of objects may be different. You have other sliders to change the aspect for instance.

I really regret it works this way but shifting does distort somewhat the geometry anyway, so the goal is not necessarily to reproduce exactly the T&S result but maybe more to get a pleasing result.

Or even a much more correct result. It's never going to be perfect,

I totally agree, it will never be perfect if you are too close to these tall buildings. T&S won't make miracles.

but there are a lot more choices to be made than a tilt shift lens can provide.

I repeat again that it works anyway with verticals and the image if the church you had posted does not proove the opposite.

I certainly prefer to reproduce the T&S result, it is in my opinion the most correct one. I this case Hugin and certainly other tools like PtGui wotk perfectly. You can post other examples if you prefer but maths don't lie. So even diagonal shift can be reproduced without difficulties.

To understand why it can be computed. Put the camera at the center of a sphere, with the sensor at distance=focal length. Place the image on the sensor and project the image on the sphere (as you can see in the Hugin screenshot I have posted). Rotate the camera at will (pitch, yaw, roll) and then project again the image from the sphere on the sensor. This is a perfect transformation.

This does not mean I consider that T&S are useless, they may be a bit better but I don't see how they could really outperform the results from software corrections. There are not horrible and unnatural results on one side and fantastic results on the other side.

Well, that's pretty much how I see it. Unnatural distortions from the tilt/shift, in this case the top corners racked to a point completely unlike what the viewer would experience at the venue.


Google Images

much improved with software:


Photoshop

Please note that you could have these results also with T&S (or close to it). by adding horizontal shift.

But you have more liberty in post processing even if, as you know, I would limit myself to the basic tools.

I consider that T&S is overkill in most cases, but that is just my opinion.

 

[Tony Bologna]

Interesting post! The two main benefits of t/s lens for me that you don't get in post are control over diffraction (I'd rather not stack images for DOF) and compositional assistance as a t/s is applied in real time while you frame vs in post at a PC. What say you?

 

[mfinley]

Today I took my 24mm TS lens to a local castle to see if there are any visual differences between using a shift lens, and correcting keystoning in post. The lens was on a tripod and in a fixed position. In one case the lens was shifted, in the other, shift was set to zero and the lens tilted to simulate a normal lens.

I looked at two cases: simple vertical shift and diagonal shift. I then made an approximate overlay of the images to get an idea of the differences.

Simple vertical shift is pretty close, with very minor differences probably due to my approximations. Diagonal shift does have some more noticeable differences.

I just wanted to do a geometrical comparison. The merits of using a shift lens compared to correcting in post are not the subject of this post.


24 TS lens shifted vertically


24mm TS no shift, lens tilted


Overlay


24mm Diagonal shift


24mm TS Lens with zero shift tilted upwards and corrected in post. 2 way correction


Overlay


24mm TS Diagonal shift


24mm TS Lens with zero shift tilted upwards and corrected in post. 2 way correcton


Overlay

“Everyone sees what you appear to be, few experience what you really are.”
- Niccolo` Machiavelli.

Please redo the test with a much greater tilt to the camera. And the best test would be trying to do this with a T/S and a non-tilt shift as T/S have better image quality characteristics to start with.

--
Thanks,
Mike

Login • Instagram

Welcome back to Instagram. Sign in to check out what your friends, family & interests have been capturing & sharing around the world.

www.instagram.com

 

[John Sheehy]

This does not mean I consider that T&S are useless, they may be a bit better but I don't see how they could really outperform the results from software corrections. There are not horrible and unnatural results on one side and fantastic results on the other side.

Pixel resolution probably has something to say, too. A 12MP FF camera probably benefits more from getting it right in-camera than a 60MP FF.