Stitching - An Advanced Approach

[BernardLanguillier]

Hi Kirk,

Sure, why not try PS if you own a license already? It is likely to do a perfect job for your applications.

Cheers,
Bernard

[BernardLanguillier]

Thank you Bernard. Many good tips that I were unaware of.

I have stayed away from getting a proper GPU before but may be getting into more video and when stitching also can benefit I might have to upgrade the computer. Any recommendations on GPUs? What to look for?

Hi Torbjörn,

Any recent GPU offering OpenCL support in your OS of choice should work I assume, but you may want to check with the targeted software support team whether there is any known issues.

I am using the Dual D700 cards of my Mac Pro 2013.

My pleasure.

Cheers,
Bernard

[kers]

hello Bernard,

Thank you for pointing me to the RRS equipment- at the moment i use my own RRS-panorama head made from RRS components,  but this one looks very sturdy and fast. Also it eliminates extra connections that could lead to less precision.
I yesterday did some test to see if a 45mmPCE  using its full circle has about the same quality as the stitched Sigma art 50mm... and Sigma won hands down...
So i can say i believe in your approach although not everything can be stitched...

about using ptGUI i have some questions...
PtGui calculates the lensdistortion... ( i always make a 360 sphere to exactly calculate the nodalpoint-distortion etc...)
Do you work from a former calculated lensprofile or just let you ptGUI go from scratch
and on stichtching - what blending method and interpolater do you like best... i have noticed the lanczos 16 has the better details...but takes longer..
for blending i use the standard ptGUI blending

[Ligament]

Bernard,

Thank you for the excellent article and sharing your hard earned knowledge on the subject.

For landscape and seascape purposes I'm nervous. It seems to me even a slight breeze would result in ghosting artifacts of branches, grasses. Not to mention the issues with water and waves.

This would not be much of a concern if one shot few panels with a lower res sensor. But I would consider stitching primarily to generate a very high res image with a D810 for example. I would assume this would magnify ghosting artifacts.

Clearly you have come to terms with this issue. I suppose my question is when you rule out pano as an option due to potential ghosting/subject movement?

Secondly, I generally maintain very strict image stabilization techniques; heavy RRS tripod weighted with my backpack, wireless remote, mirror up with EFC, sandbag on camera if needed. This would not be possible with a handheld stitched image. Based on your images I'm sure you take the same precautions when possible. How and when do you decide forgoing such stabilization is worth the blur/benefit ratio?

I hope my questions make sense and again thanks for the awesome (and free) article.

[BernardLanguillier]

about using ptGUI i have some questions...
PtGui calculates the lensdistortion... ( i always make a 360 sphere to exactly calculate the nodalpoint-distortion etc...)
Do you work from a former calculated lensprofile or just let you ptGUI go from scratch
and on stichtching - what blending method and interpolater do you like best... i have noticed the lanczos 16 has the better details...but takes longer..
for blending i use the standard ptGUI blending

Hi there,

On distorsion I just let PTGui do its magic, or sometimes use DxO.

I use PTgui stitcher and blender and typically Lanczos2 (Sybc16), sometimes higher.

Cheers,
Bernard

[Patricia Sheley]

(Off-topic) Deep felt wishes of calm and safety for your community in the love of nature. At this time of Ontake, its winds reminding us again how fragile our place here. Glad to see you posting, p.

[BernardLanguillier]

This would not be much of a concern if one shot few panels with a lower res sensor. But I would consider stitching primarily to generate a very high res image with a D810 for example. I would assume this would magnify ghosting artifacts.

Clearly you have come to terms with this issue. I suppose my question is when you rule out pano as an option due to potential ghosting/subject movement?

It depends on the subject basically. I would not attempt to stitch a forest in a storm.

But I may attempt to stitch the edge of a forest if I can capture the tree in a single image.

Secondly, I generally maintain very strict image stabilization techniques; heavy RRS tripod weighted with my backpack, wireless remote, mirror up with EFC, sandbag on camera if needed. This would not be possible with a handheld stitched image. Based on your images I'm sure you take the same precautions when possible. How and when do you decide forgoing such stabilization is worth the blur/benefit ratio?

I always do all that when I go somewhere to take pictures. Unfortunately (or fortunately ? :-)) that is a lot less frequent recently because of family constraints which forces me to stitch handheld at whatever ISO is required to get sharp images.

Cheers,
Bernard

[Rhossydd]

An interesting article.

Thanks to Bernard for piquing my interest in stitching again and going to look at the latest version of PTGUI. I last upgraded to version 5, but the improvements in version 10 are staggering, so much faster with so few artefacts. I'll be re-engaging with this now.

[Ligament]

Many thanks Bernard!

It depends on the subject basically. I would not attempt to stitch a forest in a storm.

But I may attempt to stitch the edge of a forest if I can capture the tree in a single image.

I always do all that when I go somewhere to take pictures. Unfortunately (or fortunately ? :-)) that is a lot less frequent recently because of family constraints which forces me to stitch handheld at whatever ISO is required to get sharp images.

Cheers,
Bernard

[BernardLanguillier]

(Off-topic) Deep felt wishes of calm and safety for your community in the love of nature. At this time of Ontake, its winds reminding us again how fragile our place here. Glad to see you posting, p.

Indeed Patricia.

Cheers,
Bernard

[Fine_Art]

Thanks Bernard, for a very good article.

I have had good results with Hugin, a free program. If you can't find a seam even when you know where to look the program is good enough. The only time it has failed me is with very long FLs that it does not seem to know how to shape. Im talking 1200mm. So anyone doing an occasional stitch may want to look at that program.

I have been doing the FL row stitch as well, for a while, with good results. Typically foreground at f8, mid at f8, sky/infinity at f5.6. I am impressed by your speed. My question is what do you consider the tradeoff on MLU? I typically use it, which slows me down a lot. Do you feel that tripod/ head is good enough that MLU is no longer worth the trouble? If I could do the 1ev brackets for each tile in continuous high drive mode it would make a big difference in fast changing light. Any thoughts on the topic (from others as well as our expert Bernard) would be appreciated.

[Fine_Art]

A couple more thoughts on the subject:

There has always been advise to lock the exposure, I have never done that, preferring to ETTR. The software seems to adjust the expose fine. This is a real issue when the sky is at a vastly different exposure to the landscape. What do people think with modern software?

I also want to agree with the above posts regarding getting more use from fewer lenses. I like the wides I get from the 35 stitched. I also like the 85mm. Two good lenses get me a lot of mileage.

[BernardLanguillier]

Team,

Another sample for your review. The image below was shot with the new 400mm f2.8 E FL on the D810. It is a 600 megapixel stitch that is hard to automate because of the limited DoF and many overlapping features. I used ISO64 and electronic First Curtain.



I tried this with PS CS6, PTgui Pro 10 and Autopano Pro. In the end, I was only able to get a workable image with PTGui 10 thanks to its PS layered output and close initial proposal. The layered PSB file is 14 GB. PS came up with a 2.5 gigal pixel canvas with orphane images all over the place and a large number of stitching errors in the area it could stitch.

This is one image where using a robotic head would have a huge advantage because of the perfect repeatability of the rotation increments. The grid feature of PTgui would make the pano job a breeze. Food for thought.

Another one in the same series.



Cheers,
Bernard

[dreed]

I'm curious - what do people usually do with the sources images for panoramas?

Do you keep them around in case the software improves to generate better copies?
Or once you've got an output file that you're happy with, do the sources get deleted?

[Bart_van_der_Wolf]

I'm curious - what do people usually do with the sources images for panoramas?

Do you keep them around in case the software improves to generate better copies?
Or once you've got an output file that you're happy with, do the sources get deleted?

Hi,

I don't delete source files (Raws). Raw conversion quality may improve over time, in fact it has improved. Blending engines also improve. Should the need arise, a new stitch can be made (usually with the same script file as create initially), which is simple if only the reconverted source images are changed.

It is not necessary to keep the individual TIFF tiles around, because they can be regenerated if necessary. Only if there may be tweaks required to the TIFF input images, e.g. White Balance or exposure or to the required output dimensions or projection, I might keep them around for a while.

Cheers,
Bart

P.S.  This is also why a very good pano stitching software is so valuable. When the stitching/blending produces good results that do not require additional manual intervention, life becomes so much easier. When a lot of manual work needs to be done in beating the individual warped images into submission, it becomes less likely that a new attempt to create the improved pano from scratch will ever happen.

[BernardLanguillier]

I'm curious - what do people usually do with the sources images for panoramas?

Do you keep them around in case the software improves to generate better copies?
Or once you've got an output file that you're happy with, do the sources get deleted?

I typically keep the raw files and delete the tiff generated to build the pano.

Cheers,
Bernard

[Bart_van_der_Wolf]

Has the need arisen?

Sure. Over time one may develop a slightly different taste, or the source images can be improved before blending them with the pano stitcher. For example, if image tiles have longitudinal chromatic aberration, or high ISO noise, and a better way of removing it presents itself, then I'd redo the pano.

Another example is the improvements that the 'relatively recent' (I've been stitching since the film days, with print collages and later with scanned film files) Topaz Labs 'Clarity' and 'Detail' can add to an image, which may be enough of an improvement to redo old panos.

Another one is when Capture One version 7 was released, the Raw conversion quality made a significant jump ahead. Some panos that were intended to be output large, or got improved tonality, were redone from new TIFFs, with the same stitching project settings.

Because re-using a project-file is so effortless, it is done quicker than it takes to produce the actual output.

Cheers,
Bart

P.S. There are also ongoing developments in Blending engines, that may create better output straight from the stitcher, thus allowing to replace layered Photoshop files.

[BernardLanguillier]

I thought I'd share two recent example combining DoF stacking and stitching.



D810 + Otus 55mm f1.4.



D810 + Leica 180mm f2.8 APO.

Cheers,
Bernard

[Bart_van_der_Wolf]

I thought I'd share two recent example combining DoF stacking and stitching.

Hi Bernard,

The result is, an immersive experience ...

Cheers,
Bart

[BernardLanguillier]

Another one from the same series, same technique, 200 megapixels.



Cheers,
Bernard