My Autopano Pro forum

I think these two images meet those criterion.

Hi Alexandre!

here we are: . . . . well, the upload-slots didn´t work. Both are under the size limits.
I´ll put it on my server.




best, Klaus

Last edited by klausesser (2008-03-01 13:50:19)

Yes, please.

kalain it would be better if you'd edit that post and put those image URLs into

 [url][/url]

, that way they won't load every time someone just looks at this page and you won't have to pay for wasted bandwidth.

Last edited by DrSlony (2008-03-12 16:58:01)

Hi Alexandre,

Do you want a really badly shot (handheld, lots of parallax) 360x180?

OK, they are on their way.  The images were taken at Kennedy Space Center's Rocket Garden.  The sun was bright, and I stood in the shadow of a rocket for each pano.  One pano is a 360 x 180, the other is a 360 x 90, the top half of a sphere (who needs lots of pictures of the sidewalk?) 

To see what I did with APP and a lot of editing go to http://www.hankkarl.com/KennedySpaceCen … enter.html and click on one of the top two links.  Note that I used HDView (Sorry to all the Mac and Linux guys).

The mistake I made was to make the circle centered on my feet rather than the NPP.  I went round in a circle, then changed rows.  Also, I didn't pay much attention to keeping the NPP at the same height.  So there is a lot of parallax.

DrSlony wrote:

kalain it would be better if you'd edit that post and put those image URLs into

Code:

 [url][/url]

, that way they won't load every time someone just looks at this page and you won't have to pay for wasted bandwidth.

Well, that 's what I did (with ).
I tried also without , but seems to behave in the same way.
So, I just puted in 'code'.

I'm not sure to really understand all edition possibility....

Tipjip wrote:

reason? The place! I took the photos under the Eiffel Tower. Right in the middle. I guess the stitchers got really confused with four pillars all looking exactly the same...

You can see an early version (photoshopping yet to come) at http://dreiwolf.eu/fotos/Eiffeltower.mov.

I'm uploading the photos (40 of them, converted to jpg) to http://dreiwolf.eu/fotos/Eiffelturm.zip right now. Should be ready in a few minutes, if you need them.

Regards,

Christian

I reckon this scene was always going to be very difficult to stitch when shot with a lens as long as this.

If you knew (from your pano head) the Yaw, Pitch and Roll data for your images manually entering that data may have been an alternative approach?

I think there are 42 images in your zipped set?

Tufused files, an obviously hard to stitch pair since the waves changed and the clouds are fuzzy. I uploaded to FTP a file called drslony_quay1_unstitchable.zip

For others interested, here are the same photos just rotated 90 degrees and saved as jpg @ quality ~83.

Last edited by DrSlony (2008-04-28 14:58:15)

GURL thats a good solution to bypass this problem although i think the point of this topic is to help improve the SIFT algorithm in various difficult situations, so i posted one. Yes, this was a bigger pano, there were 5 shots, 2 to the left, 1 to the right, only this pair didnt stitch automatically no matter what I tried. The funny thing is that once I manually positioned the images, i selected some similar regions, got red CPs with a high RMS, then I optimized and some of those CPs turned green... so there is definitely a chance to improve APP so it could do that by itself.

ps. ignore the fact that the CPs on the screenshot despite being green are actually bad, because they were placed on waves or clouds, the point is that no CPs were found automatically but they WERE found after I moved the images around manually - which means two things: 1) there is room for improvement of the automatic CP discovery, and 2) APP is good enough to find some CPs even in this very difficult set with a minimal amount of human help!

Hi everybody,

My first post here,
I am currently evaluating the means for so-called "2-shots" or "Double-Shot" full automatic stitching.
There are presently several and very different hardware combo solutions that allow to shoot full circular images of >>180 degrees of Angle of view. There are some software that can do automatic stitch more or reliably though. Some others can do a perfect job but from manual coontrol of the user.

From a pair of opposite such hemispherical images, the panographer can get a 360x180 panorama of more than decent quality for full-screen acceptable resolution. BTW much better than the images from a venerable CP 990 posted above by lc8b105, but with the same problem with AutopnoPro: While APP can stitch quite easily three of these circular images shot 120 deg apart (i.e. 3-shots stitch) there seems to be no way to get a decent stitch from a pair of opposite images. Furthermore APP seems to refuse to let a circular image to be of more than 180 degrees of FOV, so there is no way to manually force a good manual 2-shots stich even if APP can sometime detect correctly some CP in a crescent of the circular image.

Am I missing something?

There are some sample images on the Sunex Superfisheye web page that can be used for test. I could provide others and from different combos.

Edit (May 14,2008):
From more recent testing I have found that APP 1.4.2 can do very fine and efficient detection of CP on 2-shots images. On the same images it is also easy for APP to detect more CP by forcing it to do so when needed.

In fact it all depends on the images overlap quality and surface area. In short: the more the FOV, the best is the detection.

For example (CP P5100 + FC-E8 with an adequately modified lens adaptor) combo is providing 2-shots (190 deg FOV) with more than 95% success rate of auto detection by APP, providing that there are sufficient subject features in the images overlap region.
Same successful outcome with (EOS 5D + "shaved" Nikkor 10.5 mm) or (EOS 5D + "shaved" Tokina 10-17 mm) combos.  Cropped circular- 2-shots images must then be shot in lanscape mode, of course.

Both the Sunex 5.6 mm superfisheye and the Sigma 4.5 mm (185 deg FOV on APS-C DSLR) yield almost never any CP automatic or forced-manual detection by APP at all.

Michel

Last edited by enbilaman (2008-05-14 11:58:44)

2 shots stitching ... Let be honest. I don't like them !

Well, I don't like them either. And I look at them only by insatiable curiosity... until the right lens comes by:-)

Some people love them though. They have paid a high price for years to run their 2-shots system business.

The unique and often forgotten cabability of the 2-shots system is indisputable: it is the **only** way to assure **absolute no parallax** error wathever is the (stationnary) subject distance in the scene from 2 cm to infinity.
BTW I have done some for fun with incredible ease and on a snap. Of course, the quality is limited by the miserable size of the sensor: http://michel.thoby.free.fr/FC-E8_revis … som-2.html
This unique feature is why Stakhanovistic panographers make dozen of (real estate) panoramas per day. They shoot each and every rooms of the houses for sale or for rent, without gear dismounting.
The time advantage doesn't come mainly from shooting less images. Having to shoot more than two photos simply means the advantage of full "safe automatism" is lost and much more manual and intelligent action is required in the workflow.

TCA, softness and other annoyances near the seam can be fully overcome as the technology is already here. Nikon have shown the way with the Nikkor 10.5 mm then Pentax (designing the Tokina 10-17 mm) have followed. Both produce however a too large circular image for true 2-shots even on a FF DSLR.
My experiments have demonstrated that when the **useful** angle of view of the lens is >> 190 deg and when the projection radial mapping of the fisheye is OK, there is almost no observable loss of quality near the seam. Necessary corrections prior to stitching could eventually be made on board the camera in the near future.

I don't expect that a ~30 x 30 mm 16-20 MP sensor camera to be affordable in the near term but I am convinced that a sharp 8 mm f 2.8 fisheye with a truly useable FOV >190 deg. shall be available soon. And then, high quality (with today's standard) 2-shots panoramas shall also be at last possible.

Even with the sunex superfisheye which has a large fov, I found that it is just not the right way to work to get a good panorama ( even with other softwares which supports more this kind of stitching )

IMO Sunex has failed to invent the right lens. The "Superfisheye" can support 3-shots at low resolution with low end DSLR but I think there are other better means to produce 3-shots with higher quality and resolution. The FOV of the 5.6 mm lens is definitly too small for 2-shots and after a carefull study I have told them so. The same holds for the Sigma 4.5 mm.

Michel