[GUB]

I think it is simplest just to think about focal length (field of view) when thinking how hard it will be for the stitcher to handle field distortion. My Pentax 28mm is a rock solid lens for stitching on my K01 (apsc) but the extra fov distortion on my K1 makes it struggle at times and I should probably use my 40 pancake. Hugin had a bit of a struggle today with this 3 x 28mm shot (Full frame - freestyle with modest overlap). It took several restarts to get a working stitch.

Attached Images

View Picture EXIF

File Date

 

[Kobayashi.K]

I think it is simplest just to think about focal length (field of view) when thinking how hard it will be for the stitcher to handle field distortion. My Pentax 28mm is a rock solid lens for stitching on my K01 (apsc) but the extra fov distortion on my K1 makes it struggle at times and I should probably use my 40 pancake. Hugin had a bit of a struggle today with this 3 x 28mm shot (Full frame - freestyle with modest overlap). It took several restarts to get a working stitch.

I can't remember anybody complaining about MS ICE, and Hugin is a mixed bag. So if you have issues with Hugin simply try ICE on the same set of photos for a comparison. It's free and easy to use.

[GUB]

So if you have issues with Hugin simply try ICE on the same set of photos for a comparison. It's free and easy to use.

Naaa I think of it as Hugin having issues with my efforts!

[tokyoscape]

My 2 cents, I stitch number of images and using mainly DA21, FA50 macro, DA*55, DFA100wr. Give it at least 30% overlap and pre-corrected images using Adobe lens profile fist. Then let CameraRaw do the Merge to Panorama and save it as DNG. Do another post-processing after that. The limitation is my computer. If it does more than 10 images at one It will run very slow.

Images from FA50 macro and DFA100wr can have less than 30% overlap and still need no pre-correction.
sample,
DA15 - 3 images stitch
FA20 - 3 images stitch
DA21 - 3 images stitch
F28 - 3 images stitch
DA*55 - 3 images stitch
FA50 f2.8 - 3 images stitch (this is the lens I feel sorry letting it go.)
100wr - 6 images stitch (just sell this lens and I start to feel sorry now!)

---

Last edited by tokyoscape; 02-19-2020 at 04:07 PM.

[travelswsage]

Wonderful images tokyoscape -- especially love the one at night from the Fa 50.

[JensE]

That's a good thing, pre-corrected images helps the stitching software to only have to deal with perspective distorsion.

I found that to not be the case with Hugin. Hugin normally calculates its lens distortion model from the overlap of the images and optimizes the overall fit by tuning the distortion model. At least for lenses with a relatively simple distortion geometry, it always seems to work better than correction distortion upfront. The correction of the significant deviation from the claimed stereographic projection of the Samyang fish-eyes (8mm APS-C, 12mm FF) also works really well.


Even the 0.07% of distortion of the DFA100WR still translate to ~3 pixel in the corner of a 36MP image, causing significant blur if not smoothly blended in the transition zone. So letting Hugin optimize its model for the best match of well-chosen control points will usually be better than an approximate generic distortion model, potentially measured at a different focal distance than your image was captured.

I really need to try stitching with the 50mm and 100mm macro lenses, never thought of it. I'm usually stitching more problematic shots, like the D-FA 28-105mm close to the wide end, when I don't have anything wider with me.

[yucatanPentax]

My 2 cents, I stitch number of images and using mainly DA21, FA50 macro, DA*55, DFA100wr.

All of your images are wonderful. It's tempting me to try my own panorama stitching after not doing it for some time. The image with the FA50 is especially nice, but I'm also impressed by the DA15 pano. All are very nice, but to me, those two really stand out somehow.

[biz-engineer]

sample,DA15 - 3 images stitchFA20 - 3 images stitchDA21 - 3 images stitchF28 - 3 images stitchDA*55 - 3 images stitchFA50 f2.8 - 3 images stitch (this is the lens I feel sorry letting it go.)100wr - 6 images stitch (just sell this lens and I start to feel sorry now!)

Oh yes. You have here fantastic stitched panoramas. You managed to stitch from 20mm, it's wide already. It could be that the many features and lines contained in each frame of city secenery help for stitching. When stitching indoor archi, there can be entire areas of frame without detail, making it harder to align frames relative to each other.

---------- Post added 20-02-20 at 08:37 ----------

Hugin normally calculates its lens distortion model from the overlap of the images and optimizes the overall fit by tuning the distortion model.

Sure, hugin documentation also says so. I've read that some kind of lens distorsion ("moustache"?) can't be corrected by the a,b,c model, and I was assuming that a lens having significant distorsion might also have some of that distorsion not fully corrected by the a,b,c model, leading to mismatches between frames.

---------- Post added 20-02-20 at 08:40 ----------

You can clean up an image until the result conforms to the idea of an end-result you have in mind.

Agreed. At the end, it's the creation of the final image that counts, depending of photographer's intent. That's precisely how fine art photography is defined, bringing the photography creation process a bit closer to the painting process.

[JensE]

QuoteOriginally posted by JensE QuoteHugin normally calculates its lens distortion model from the overlap of the images and optimizes the overall fit by tuning the distortion model.Sure, hugin documentation also says so. I've read that some kind of lens distorsion ("moustache"?) can't be corrected by the a,b,c model, and I was assuming that a lens having significant distorsion might also have some of that distorsion not fully corrected by the a,b,c model, leading to mismatches between frames.

That's why my reference to relatively simple distortion. The good thing is that, dependening on how you shoot and what your subject is, it may not even matter. For stitching, the model only needs to be very accurate for the area of overlap. So if you e.g. shoot a single row panorama at constant angles per shot from a tripod and overlap by, say, 1/5 of the frame on either side (or even handheld using the grid), the remaining 60% don't come into play, so that a central 'moutache dip' doesn't hurt if it is not visually prominent through architectural line or such. Those conditions happen to be true for most of my landscape panos.

Otherwise, it's of course OK if the residual un- or overcorrected distortion can be approximated well enough by Hugin's (a,b,c) model - a rotary symmetric cubic polynomial in the radius. According to https://download.macromedia.com/pub/labs/lensprofile_creator/lensprofile_cre...ameramodel.pdf, there is quite a different model in use for Adobe profiles, especially also containing a tangential model in first order. So the residual may not fit the radial cubic model very well ...

For now I'll stick with my practical experience but will give pre-correction another try e.g. on the 14mm Samyang. I wish I had "a little time" to get into code and experiment with some models ... But my limited spare time is better spent on taking better pictures.

[biz-engineer]

Perspective distorsion and lens distorsion are the main enemies of lines when stitching multiple images. Non line features of an image are much less constrained by spherical projection.
How about identifying lines in a composition and organizing frames around those lines to avoid having lines cut between frames?