Learning Thread � Landscape Photography

06/01/2006 11:55:13 PM · #726

Originally posted by fracman:

Ok, I finally did my own comparisson of the tools I've got...

Autopano Pro
Autostitch
Photomerge (CS2)
Hugin

Trying to look at the large sized images but it is taking a month of sundays for then to show on my PC. I might be abusing my RAM. ;)

I did notice that both Autostich and Autopano Pro messed up basically in te same plcae and that there was some lesser merge issues in the trees on the horizon almost directly above.

The Photomerge version just plain sucks! :) How come that version has more detail on the right side? That is even worse than what I've experienced. Btw... I used Photomerge for that large 3 row and 30 image Cathedral Rock pano. I fixed all thos issues, though it seemed to me that mergewise it performed better than with your images.

Having trouble bringing up the Hugin image... think I will reboot and see if that helps things all the way around. :)

06/02/2006 12:08:21 AM · #727

Steve and Jon,

I've found your comparisons of the stitching software to be very interesting. Both of you are showing serious problems with the stitching tool included with Photoshop. I've been more successful with it:

Was I just lucky?

How much difference does it make that I am using a pano head for my tripod that lets me rotate the camera around the nodal point of the lens? I have the pano head from Jasper Engineering.

My Ohio River flood image was done before I got CS2 so the software was the earlier CS version.

I hope to find time to shoot and process a new pano in the next few days. I'll let you know how it turns out.

--DanW

06/02/2006 12:10:32 AM · #728

Originally posted by stdavidson:

Found a Pano software just updated just yesterday! Panorama Factory:

Not to be outdone, I grabbed the demo of Panorama Factory and ran the same set through it.

Panorama Factory processed pano

On the plus side, I much prefer the interface of this tool to pretty much all the others. The part I liked the most was in the manual control point screen where it gave an indication of the quality of the points I was placing. The ability to temporarily zoom to 500% (when it worked) was also nice.

On the down side, while it was able to stitch all 7 frames, there were visible seams at all stitch points. The seams weren't as noticable as the Photomerge ones, and thanks to the complexity of the image, the ghosting was further minimized, but overall, I think I prefer the output from Hugin.

Message edited by author 2006-11-14 16:20:00.

06/02/2006 12:22:13 AM · #729

Originally posted by wheeledd:

Was I just lucky?

How much difference does it make that I am using a pano head for my tripod that lets me rotate the camera around the nodal point of the lens?

Just lucky? Not at all. I can't speak for Steve, but I know that the lack of the correct hardware (i.e. an actual pano head) is the cause of a significant amount of my problems. To be honest, with the excpetion of my Badlands pano, I can't think of a worse case for testing than the series I shot at the dam tonite. There are very few straight lines, I only left maybe 100 pixels overlap between frames 5 and 6, I handheld the whole series, pivoting on the vertical axis of my body rather than the nodal point of my lens, and I accidentally used the wrong mode when shooting, which led to a re-metering of the scene on frame 4 (I think). Add to that the moving water and the fact that the two fishermen didn't want to stand still while I shot, and you've got a pano series that'll bring the best to their knees :-)

I found some instructions on making a pano head that I'm going to try some day. I can't find the right link at the moment, but I'll post it when I find it again.

Canon EF 17-40mm f/4L USM

06/02/2006 12:35:51 AM · #730

A side note to Rikki:

While looking for my earlier pano shots, I came across some snapshots of John Paul Caponigro that I took a couple years ago at a Maine Photographic Workshop:

click:

Most of the images in the album were taken at his gallery/house. The others show workshop participants at our dining tent.

BTW, although he certainly does heavy postprocessing, very little of what I saw has the hyperprocessed look that we have been discussing in this forum. What he taught us to do was much more subtle.

--DanW

Rikki

Canon EOS-20D

06/02/2006 12:43:45 AM · #731

Ya I agree. His saturated images are different from the ones we refer to as hyper-processing. I just took it a step further... just a little notch :)

Samsung Galaxy S21, S21+, S21 Ultra

06/02/2006 12:44:13 AM · #732

Originally posted by fracman:

I found some instructions on making a pano head that I'm going to try some day. I can't find the right link at the moment, but I'll post it when I find it again.

I've also seen some instructions recently but I've forgotten where. I wasn't very interested because I already have a pano head. Steve's multirow panos have made me want to try some of my own, but my pano head does not do the up-down tilt on the nodal axis. My gut feeling is that it is probably even more important in stitching multirow images.

Steve, what do you do? Are you careful to rotate about the nodal axis in both dimensions?

--DanW

Bear_Music

Sony 24-70mm F4 Vario-Tessar T* FE OSS

06/02/2006 12:48:45 AM · #733

Originally posted by Rikki:

Ya I agree. His saturated images are different from the ones we refer to as hyper-processing. I just took it a step further... just a little notch :)

That's our Rikki, the Emeril of photography...

R.

06/02/2006 12:51:48 AM · #734

Ok... OK... I broke down and did a Hugin merge of St Helens "by hand". Hate that. Here it is:

//www.pbase.com/azleader/image/61167657

I decided to do that because I did it with PanoFactory and had to go over and over the control points so many times I'd thought I remembered that Hugin was much easier than that. It is. It sucks into the matched points nicely when doing things "by hand" and is about 10 times better than PanoFactory.

I only put in a few control points so, as I suspected, the merge wasn't all that great. You can even see it in the smaller image I posted plainly in three places. The blend did not seem as good either, but it did use Emblend. Still the blending is so much better than Photomerge.

Another annoying thing I rediscovered about Hugin is that it produces a truckload of intermediate files that it does not clean up after it is done.

Does that happen to you, fracman, when you build panos or is that symptomatic of things not working right on my machine?

If I could only get SIFT to work, I'd be happy. :)

Message edited by author 2006-06-02 00:53:17.

Canon EF-S 18-55mm f/3.5-5.6 IS

06/02/2006 12:59:14 AM · #735

Originally posted by wheeledd:

Originally posted by fracman:

I found some instructions on making a pano head that I'm going to try some day. I can't find the right link at the moment, but I'll post it when I find it again.

... Steve's multirow panos have made me want to try some of my own, but my pano head does not do the up-down tilt on the nodal axis. My gut feeling is that it is probably even more important in stitching multirow images.

Steve, what do you do? Are you careful to rotate about the nodal axis in both dimensions?

I did NOT have anything special for my 3 row panos. I just mounted my camera on a standard tripod. However, my 3 row panos are only about 120 degrees at most. It might make a bigger difference if it is wider. One of the narrow panos I did was a little over 180 degrees but it was taken on a standard tripod as well. I think that keeping the tripod level and having 25-33 percent overlays is as much a key to good panos as a pano head would be, but that is just a guess.

justin_hewlett

Canon EOS-1000D Rebel XS

06/02/2006 01:03:04 AM · #736

Originally posted by stdavidson:

In my experiments stitching together panoramas seems to have two main challenges:
1-The merge itself
2-Luminosity(brightness) blending

I've done a bit of manual stitching in PS. I found a workable solution to the luminosity problem - and it was something along the lines of adjusting the levels on the Red, Green, and Blue channel B/W images individually. Then, as soon as all three B/W channel images matched up, you could be sure that the color image would match up as well.

I can't remember the exact process but I just remember it had to do with matching the gray tones of the Red, Green, and Blue channels individually. Maybe Rikki or another PS guru can fill in the details.

06/02/2006 01:13:38 AM · #737

Originally posted by justin_hewlett:

Originally posted by stdavidson:

In my experiments stitching together panoramas seems to have two main challenges:
1-The merge itself
2-Luminosity(brightness) blending

I've done a bit of manual stitching in PS. I found a workable solution to the luminosity problem - and it was something along the lines of adjusting the levels on the Red, Green, and Blue channel B/W images individually.

Believe me, I know... I've done a bunch of PS panos and did a lot of that stuff. My approach was to treat it like a color cast and fixing the blends that way along with judicious cloning. The problem isn't so much correcting the color as it is making a decent selection with a proper feather. No matter what it is a major pain. I don't think PS should be used to make panos, it just does a poor job. It does a poor job with noise reduction to and that is why we need to use third party add-ons like NeatImage.

06/02/2006 01:22:04 AM · #738

Originally posted by fracman:

Originally posted by stdavidson:
Found a Pano software just updated just yesterday! Panorama Factory:

Not to be outdone, I grabbed the demo of Panorama Factory and ran the same set through it.

Panorama Factory processed pano

On the plus side, I much prefer the interface of this tool to pretty much all the others. The part I liked the most was in the manual control point screen where it gave an indication of the quality of the points I was placing. The ability to temporarily zoom to 500% (when it worked) was also nice.

Though I liked the quality display on control points I thought the interface for "Hand" setting control points is much crappier than Hugins. I also had to readjust a lot of point 3 or 4 times to get any quality at all out of them. Hugins sucks the control points together when you get close and I like that, Panofactory doesn't. At 500% it was harder for me to move the control points aroung than in the 300% or whatever size is used in Hugin.

PS. You notice how screwed up the rocks in the foreground were with that merge. Must be you didn't get enough good match points there or something. The merge at the the top looked pretty darned good to me.

Message edited by author 2006-06-02 01:28:41.

06/02/2006 07:45:27 AM · #739

Originally posted by stdavidson:

Another annoying thing I rediscovered about Hugin is that it produces a truckload of intermediate files that it does not clean up after it is done.

Does that happen to you, fracman, when you build panos or is that symptomatic of things not working right on my machine?

As Alice would say, curiouser and curiouser... Nope, the only program that left a bunch of intermediate files was Pano Factory. Hugin may do intermediary files, but it appears to clean them up for me at the end.

06/02/2006 08:01:06 AM · #740

Originally posted by fracman:

I found some instructions on making a pano head that I'm going to try some day. I can't find the right link at the moment, but I'll post it when I find it again.

Found the links. Two sites, one looks easier to make, but not as nice. The other, obviously, is the inverse:

The easier one
The other one

I got a buddy with a machine shop, so I'm going to ask him for a hand and, hopefully, get the more complicated one made.

06/02/2006 10:42:20 AM · #741

Originally posted by Steve to Jon:

PS. You notice how screwed up the rocks in the foreground were with that merge. Must be you didn't get enough good match points there or something. The merge at the the top looked pretty darned good to me.

I'm speaking from theory, not from experience. Being careful about rotating around the nodal point doesn't matter much for distant landscapes because you get virtually identical views from camera positions that are anywhere within a few inches of each other. But when you shoot something close, camera position becomes critical. There is no mathematical transformation that will bring two shots into agreement unless they were shot through the same nodal point.

Is it possible that Jon's foreground rocks were close enough so that the merge problems come from the changes in the camera view? Would these merge problems go away if he used a pano head? I don't have the experience to know how close you have to be for the nodal point rotation to become important.

--DanW

06/02/2006 10:47:58 AM · #742

Originally posted by fracman:

Hmmmmm... Sounds like another mystery from the geek world. :)

On the plus side, I figured out my problem with Hugin and it is working just fine. I was pointing Hugin to the "wrong" autopano.exe file. (with my tests and such I have at least three now). After pointing it to the right one things worked.

I've done three Hugin tests now.

Test 1 was with all "hand" generated control points. I did about 5-6 pairs per linked image pair. Test 2 was with 20 autopano generated control points for each pair. And test 3 was with 20 autopano-SIFT generated control points for each pair. (SIFT is supposed to be better)

The results are interesting. My "hand" generated control points, even though there were fewer, gave the best merge. The autopano generated control points gave the worse merge of any pano software I tried out, including Photomerge. And the SIFT generated control points did a lot better than the autopano ones, but still had a remaining merge issue that would have to be corrected.

Conclusions regarding Hugin and merge issues:
Select and use autopano-SIFT to automatically generate about 20 control points or so per image pair. Then go through and look at the control points and add one or two really obvious ones per image pair that SIFT overlooked. I bet the merge turns out pretty darned good!

06/02/2006 10:51:56 AM · #743

Originally posted by fracman:

The other one

I got a buddy with a machine shop, so I'm going to ask him for a hand and, hopefully, get the more complicated one made.

This one looks really good. He's taking panoramas of insides of rooms and doing multiple rows. For this kind of work, precise rotation about the nodal point is critical. I'd like one of these. The pano head I bought only handles horizontal rotation. This one looks like both horizontal and vertical rotation is around the nodal point.

I think he left out an important part of the instructions: how did he determine the nodal point of his lens and then design the head so that rotation in both directions was around the nodal point?

--DanW

06/02/2006 01:25:32 PM · #744

Originally posted by stdavidson:

Conclusions regarding Hugin and merge issues:
Select and use autopano-SIFT to automatically generate about 20 control points or so per image pair. Then go through and look at the control points and add one or two really obvious ones per image pair that SIFT overlooked. I bet the merge turns out pretty darned good!

Hmmmmm... Thought I'd try out my own theory about Hugin merges on a Pano I previously did with Photomerge but was not happy with either the merge or blend.

Comparison:
Old Photomerge (CS2) result AFTER fixing errors and doing some post:
//www.pbase.com/azleader/image/55220115

Hugin SIFT + "hand" added control points without post:
//www.pbase.com/azleader/image/61190088

I'm somewhat disppointed in the Hugin result. The blend is not very good and I was unable to eliminate a noticeable merge defect on the post after additional control points added by hand specifically to reduce it. It is notable that the merge defect on the post is not on an overlapping area of two images so control points could not be put there. Blending caused the merge defect.

Hugin settings:
Merge engine "nona", equirectangular, Calculated panorama angle used, maximum size calculated and output set to .tiff with "soft blending" using enblend turned on. If enblend turned off then I get blending as bad as Photomerge's but the merge is perfect! :(

Comments?

PS: Did quick Autostitch of this one:
//www.pbase.com/azleader/image/61192533

Better merge, better blend but more distortion. Will the fun never end? LOL!

Message edited by author 2006-06-02 14:16:34.

06/02/2006 01:51:40 PM · #745

Originally posted by stdavidson:

I'm somewhat disppointed in the Hugin result. The blend is not very good and I was unable to eliminate a noticeable merge defect on the post after additional control points added by hand specifically to reduce it. It is notable that the merge defect on the post is not on an overlapping area of two images so control points could not be put there. Blending caused the merge defect.

Would it be possible to get access to the original frames? I'd like to make a run at it and see if I get different results than you.

Jon

06/02/2006 02:18:36 PM · #746

Originally posted by fracman:

Would it be possible to get access to the original frames? I'd like to make a run at it and see if I get different results than you.

Good idea... PM me an email address and I'll send you a .zip with the original files.

Note: Hmmmm... .jpg files are about 10.9 Megabytes. Probably to big to email. Can email 1 at a time though.

Message edited by author 2006-06-02 14:23:22.

06/02/2006 02:58:36 PM · #747

OK... The files are in the mail. I'll be interested to see how your pano turns out. I did it pretty generic for the most part, but you and I could be using different parameters in our Hugin runs that may make them turn out different.

Hope mine are wrong, I'd like a better blend than I'm getting. :)

06/02/2006 04:41:16 PM · #748

Originally posted by stdavidson:

Got the pics and ran them through Hugin quick... The autopano-sift settings were: 20 points per overlap, downscale resolution 1300, automatic pre-aligning of image with bottom is right, and enable revinement step. Once APS was done (83 points autogenerated), I manually added 2 points on the tall gate post between frames 2 and 3. I then ran the optimizer for positions, view, and barrel. On the stitcher, I calculated FoV and optimal image size, used the PTStitcher with poly3 (bicubic) interpolator, exposure correction on brightness and color with the exposure anchor on image 2 (really 3, but offset is 0), and no approximation.

By my eye, the unedited version has no visible stitching problems and the luminosity looks fairly even across the scene. The whole thing looked a bit flat, so I did some quick PP and got this image. Sorry I couldn't post the PP image full resolution, Gallery2 simply barfs at 12MB files!

So, it looks to me as if those settings worked fine. I did get the messages about cropping on your input images... maybe something about the Sony EXIF headers?

Message edited by author 2006-11-14 16:19:08.

06/02/2006 05:25:50 PM · #749

Originally posted by fracman:

Originally posted by stdavidson:
OK... The files are in the mail. I'll be interested to see how your pano turns out. I did it pretty generic for the most part, but you and I could be using different parameters in our Hugin runs that may make them turn out different.

Hope mine are wrong, I'd like a better blend than I'm getting. :)

Got the pics and ran them through Hugin quick... The autopano-sift settings were: 20 points per overlap, downscale resolution 1300, automatic pre-aligning of image with bottom is right, and enable revinement step. Once APS was done (83 points autogenerated), I manually added 2 points on the tall gate post between frames 2 and 3. I then ran the optimizer for positions, view, and barrel. On the stitcher, I calculated FoV and optimal image size, used the PTStitcher with poly3 (bicubic) interpolator, exposure correction on brightness and color with the exposure anchor on image 2 (really 3, but offset is 0), and no approximation. ...

So, it looks to me as if those settings worked fine. I did get the messages about cropping on your input images... maybe something about the Sony EXIF headers?

Yeah... looks a lot better. Thanks for sharing your settings. I will work with your settings and see where mine went wrong. One that sticks out in my mind is you used PTStitcher whereas I used Nano.

Your merge is perfect. If you look very closely you can see very faint stitch boundaries in your sky. Not bad at all, though. In fact, yours are in exactly the same place but not as apparent as in this earlier pano I did with Hugin:
//www.pbase.com/azleader/image/61190093

The merge in my earlier one was just as good as yours. The reason in that one was that "soft blend", a .tiff option, was not selected. It was the soft blend that messed up the post in my pano. That is why adding more control points did not help.