7D Stitching

[Nick Rains]

People often dismiss the APS-C sensors for wide landscape work, preferring FF cameras. It just occurred to me that using the new 17 and 24mm TSE lenses on the 7D for a left/centre/right flat stitch would give you an uninterpolated image roughly 10000x3450px, as opposed to about 8000x3700px on a 5D2. For those who like the 3:1 format this would produce a stunning 50" print @ 200dpi.

I may have to get one of these...

[BernardLanguillier]

People often dismiss the APS-C sensors for wide landscape work, preferring FF cameras. It just occurred to me that using the new 17 and 24mm TSE lenses on the 7D for a left/centre/right flat stitch would give you an uninterpolated image roughly 10000x3450px, as opposed to about 8000x3700px on a 5D2. For those who like the 3:1 format this would produce a stunning 50" print @ 200dpi.

Totally true. It has always been, and remains, my opinion that DX is inherently better suited for landscape work than FF (specially so for stitching applications). The better potential for high resolutions and slightly superior DR with FF are the only reasons why I have been using FF bodies for 2 years.

I am really glad to see that Canon keeps pushing the enveloppe with DX and hope that other vendors will follow their lead.

Cheers,
Bernard

[Jeremy Payne]

... It just occurred to me that using the new 17 and 24mm TSE lenses on the 7D for a left/centre/right flat stitch would give you an uninterpolated image roughly 10000x3450px ...

How does one do that math?

Personally, I want to understand/calculate what a D700 + the 24/45/85 PCEs can do when flat stitched in each orientation.

[JeffKohn]

How does one do that math?

Personally, I want to understand/calculate what a D700 + the 24/45/85 PCEs can do when flat stitched in each orientation.

The short answer, is that 11mm shift is a larger amount relative to the DX frame than the FX frame, so you can more resolution. This only really holds true when comparing cameras with similar resolutions though (rather than similar pixel densities).

But I have to wonder how many of Canon's lenses can hold up to the pixel density of a 17mp DX sensor. I just don't see the point in cramming that many pixels into a DX sensor (unless maybe they also dropped the AA filter).

[Ray]

The new TS-E lenses from Canon have 12mm of shift in each direction. With a 7D held vertically, and shifting a total of 24mm from one extreme to the other (horizontally), you get an equivalent sensor size of 38.9mm x 22.3mm, or 9,000 x 5180 pixels, which includes ample overlap. That's sufficient for a 21.5" x 37.5" print at 240 dpi without interpolation; just right for my 24" wide Epson printer.

Note also, that even these much improved TS-E lenses from Canon are still a bit soft at the extreme edges and corners, at maximum shift with full frame. The cropped format sensors never reach the extreme edges or corners, even at maximum shift. I would expect resolution of such a stitch to be superb from corner to corner.

[Panopeeper]

What is the advantage of the T/S lens compared to the "normal" lenses in landscaping, when one anyway plans to stitch? I understand, that it can be very useful in architectural photography, but I don't see the point with landscaping. Beside, at least for me, the proportions of the resulting pano are dictated by the landscape, not by the lens. In other words, I don't mis-frame a pano for the sake of the lens, when it needs 3.5:1 or 4:1 or much more.

[Nick Rains]

What is the advantage of the T/S lens compared to the "normal" lenses in landscaping, when one anyway plans to stitch? I understand, that it can be very useful in architectural photography, but I don't see the point with landscaping. Beside, at least for me, the proportions of the resulting pano are dictated by the landscape, not by the lens. In other words, I don't mis-frame a pano for the sake of the lens, when it needs 3.5:1 or 4:1 or much more.

It's really just another tool but it's easier to stitch as there are virtually no corrections to do. And you can use a polariser easily too.

I personally quite like the 2:1 or 3:1 ratio of a horizontal flat stitch and compose to that shape accordingly - much like shooting on a 6x17 film camera. Also, because it's a flat stitch with no rotation it is exactly like a wide lens on a bigger sensor. If you like this perspective it's good, if you like the cylindrical projection panorama then that's fine too. They are just different beasts.

[Nick Rains]

The new TS-E lenses from Canon have 12mm of shift in each direction. With a 7D held vertically, and shifting a total of 24mm from one extreme to the other (horizontally), you get an equivalent sensor size of 38.9mm x 22.3mm, or 9,000 x 5180 pixels, which includes ample overlap. That's sufficient for a 21.5" x 37.5" print at 240 dpi without interpolation; just right for my 24" wide Epson printer.

What I find interesting is that the calculation you have made, using three frames joined like that, makes a US2000 camera comparable with a US40000 MFDB. The P65+ is 8984 x 6732px, same width, just a bit taller! I'd expect more DR in the Phase but for 20 times the price...?

[Ray]

What is the advantage of the T/S lens compared to the "normal" lenses in landscaping, when one anyway plans to stitch? I understand, that it can be very useful in architectural photography, but I don't see the point with landscaping. Beside, at least for me, the proportions of the resulting pano are dictated by the landscape, not by the lens. In other words, I don't mis-frame a pano for the sake of the lens, when it needs 3.5:1 or 4:1 or much more.

Gabor,
I think the advantage would be you can rely upon getting a perfect stitch every time, without a dedicated pano head, even using a standard ball-head.

It used to be the case when using TS-E lenses for stitching, with older versions of stitching programs, if the subject was fairly close, it would be necessary to remove the slight parallax errors by moving the camera body in the opposite direction to the lens shift, using a special RRS head designed for the job.

However, I find this is no longer necessary with Autopano Pro. A perfect stitch every time, which hardly even needs cropping.

However, it's true that the desired aspect ratio for the composition will not necessarily allow you to use the maximum number of pixels, but you do have the option of stitching with camera horizontally to give you a scene of 10,000x3450 pixels, or with camera vertical for a 9,000x5180 pixel stitch. You've also got the choice of a 17mm or 24mm lens (and the older 45mm and 90mm TS-E).

You also have the option of using these lenses at their sharpest aperture employing tilt with shift in order to get an increase in effective DoF, if the subject is appropriate.

It's no wonder these lenses are selling like hot cakes   .

[Ray]

What I find interesting is that the calculation you have made, using three frames joined like that, makes a US2000 camera comparable with a US40000 MFDB. The P65+ is 8984 x 6732px, same width, just a bit taller! I'd expect more DR in the Phase but for 20 times the price...?

Nick,
Shush! We have to be careful not to upset owners of MFDBs, don't we   .

I'm already planning the photos I will take with these TS-E lenses and I don't even own any of them yet, nor a 7D of course. It's true the new 7D will not compete with the P65+ in terms of SNR and DR at its base ISO of 44. However, the 7D's fast frame rate of 8 fps makes it ideal for auto exposure bracketing. Autopano Pro does an excellent job of automatically stitching and merging to HDR in the one process.

[BernardLanguillier]

Also, because it's a flat stitch with no rotation it is exactly like a wide lens on a bigger sensor. If you like this perspective it's good, if you like the cylindrical projection panorama then that's fine too. They are just different beasts.

Both T/S and cylindrical stitching are just tools, but as I am sure you are aware, you can also get flat projections from cyliindrical stitches and the result is also exactrly like that of a wide angle.

Cheers,
Bernard

[elf]

What is the advantage of the T/S lens compared to the "normal" lenses in landscaping, when one anyway plans to stitch? I understand, that it can be very useful in architectural photography, but I don't see the point with landscaping. Beside, at least for me, the proportions of the resulting pano are dictated by the landscape, not by the lens. In other words, I don't mis-frame a pano for the sake of the lens, when it needs 3.5:1 or 4:1 or much more.

The shift doesn't add anything useful when compared to a good spherical pano head properly adjusted.  Both are capable of creating a stitched image with no errors.  The tilt, however, gives you much more control over the plane of focus.  There are currently no commercial spherical pano heads on the market that can do tilts (as far as I know).

[MarkL]

What is the advantage of the T/S lens compared to the "normal" lenses in landscaping, when one anyway plans to stitch? I understand, that it can be very useful in architectural photography, but I don't see the point with landscaping. Beside, at least for me, the proportions of the resulting pano are dictated by the landscape, not by the lens. In other words, I don't mis-frame a pano for the sake of the lens, when it needs 3.5:1 or 4:1 or much more.

I don't have a pano head or any other gear other than a levelling base to ensure flat horizons and I've never had a problem with stitching any landscape I've shot. I'll save my £1500 per lens and forgo the inherent limitations with the number of frames that can be stitched. Tilt is also moot with helicon focus being available.

If I was doing interiors or architecture it would be a different story entirely though.

[Luis Argerich]

Interesting!

I've found that I mostly use my TS lens to reduce the DOF and not to increase it. The results are sometimes surprising.
For landscapes I prefer to take several shots using live view to focus precisely where I want and then merge the shots manually or using Helicon Focus. This gives me better results and control than a TS lens tilted.
About stitching I think PtAssembler, Hugin or PtGUI can do a great job without a shift lens or a pano head in 80% of the landscapes I shoot, for the other 20% I just use a home made panohead.

So what is interesting? It is interesting how every photographer uses different tools and techniques to achieve the same results. I think all the roads are valid it just depends on what each of us prefer for one reason or the other.

[Christopher]

Interesting!

I've found that I mostly use my TS lens to reduce the DOF and not to increase it. The results are sometimes surprising.
For landscapes I prefer to take several shots using live view to focus precisely where I want and then merge the shots manually or using Helicon Focus. This gives me better results and control than a TS lens tilted.
About stitching I think PtAssembler, Hugin or PtGUI can do a great job without a shift lens or a pano head in 80% of the landscapes I shoot, for the other 20% I just use a home made panohead.

So what is interesting? It is interesting how every photographer uses different tools and techniques to achieve the same results. I think all the roads are valid it just depends on what each of us prefer for one reason or the other.

I still haven't found helicon focus to be any good. Perhaps because in many of my images the wind is to strong so that grass and trees are moving to much, but I'm very happy with all the movements my large format camera offers. Than reducing DOF is something even greater, which so far can't really be done in post.

[madmanchan]

A flat stitch, if prepared carefully, can be executed without requiring image resampling. This has a small benefit for those seeking ultimate resolution.

[Panopeeper]

A flat stitch, if prepared carefully, can be executed without requiring image resampling. This has a small benefit for those seeking ultimate resolution.

I am not sure what you mean with "image resampling". The stitching always includes warping, except if one stitches scanned images; this results unavoidably in interpolation of all pixels. Of course, the pixel dimensions of the source frames play a role, for the result of interpolation depends on the proportion of the number of source and target pixels.

This has nothing to do with how one prepares the stitching but with the location of the horizon, the angle of view, the selected projection, etc., and, of course, with the pixel dimendions of the source frames and of the resulting pano.

[JeffKohn]

I am not sure what you mean with "image resampling". The stitching always includes warping, except if one stitches scanned images; this results unavoidably in interpolation of all pixels. Of course, the pixel dimensions of the source frames play a role, for the result of interpolation depends on the proportion of the number of source and target pixels.

This has nothing to do with how one prepares the stitching but with the location of the horizon, the angle of view, the selected projection, etc., and, of course, with the pixel dimendions of the source frames and of the resulting pano.

I think Eric's point was that if you do a flat stitch in Photoshop, there is no projection or warping and hence no interpolation. This is what I do for shift-lens stitching, and the nice thing is that I don't have to worry about the center/vanishing point not getting set correctly, etc. The images just line up and that's it.

[madmanchan]

Yes, Jeff, that was indeed what I meant. Sorry for being unclear in my original post. When I said "flat stitch" I meant using the shift lens approach. No need to reproject, and hence no warping.

[Panopeeper]

Yes, Jeff, that was indeed what I meant. Sorry for being unclear in my original post. When I said "flat stitch" I meant using the shift lens approach. No need to reproject, and hence no warping.

Ok, I see. However, this can work only if the geometric lens distortion is negligable; I don't know if a 24mm commercial lens is ever capable of that. Perhaps pre-correction the distortion helps.