My Autopano Pro forum

The main problem:  GIGO

Problems:

1. zenith photo does not seem to have EXIF, so you have to tell APP its an 8mm fisheye lens.
2. images are very soft
3. images are very noisy
4. 5 of the 8 images are linked to 7 other images, the remaining 3 images only link to 6 other images--too many bad links, probably due to the noise.

Looks like you shot at ISO 400, f/8, 1/15, then pushed it a stop in PS, then increased brightness and otherwise manipulated the images, then sharpened the images _and_ the noise.  I'm surprised, I would have thought the D300 had less noise than that at ISO 400. 

A little work in the CP editor gives a fair result, which may be able to be improved.

Take home messages seem to be:

1. Don't pre-process images in a manner that strips the EXIF data.
2. Get the exposure and focus right when shooting.
3. IMO this is a good example of where you'd have been better off shooting the zenith at say +60 rather than +90 to get an easier link between than main row and the zenith.
4. Alternatively tilt up about 15 degrees (?) to shoot the main row so that you don't need a zenith shot because the main row will provide enough overlap without one.
5. Use the text and badges above to set CPs manually to link the main row images and the main row tro the nadir image.

I found it very tricky to get decent links between your zenith shot and the main row.

Do you really need 6-around in the main row when shooting with a D300 and Sigma 8mm FE - I thought 4-around would provide 360 coverage?

If I was shooting the scene with my D40 and Sigma 8mm on NN5 head I'd shoot 4-around at +5 and one up at +60 (possibly two up, each located to show a distinctive doorway) - this would leave 'hole' at the nadir slightly smaller than the central rotator. If you then want to shoot a nadir you can patch it in after stitching. I'd not bother and either limit FOV on the player or place a cap at the nadir.

How did you shoot/produce the nadir image with having yourself in the shot - perhaps it is already heavily post-processed?

Anyway, you can download my best effort (so far) from here, not perfect but quite good IMO, this is straight from APP with no post-processing:
http://www.three60views.org.uk/images/t … e_pano.jpg

And here's what the links looked like prior to rendering using Spline36 interpolator and Smartblend.

Last edited by mediavets (2008-07-22 16:06:49)

Here's my best effort as a 360x180 Flash VR - reduced size, adjusted levels, despecekled and sharpened a little after stitching, then Flash VR produced Pano2VR.

It is not perfect but it's not too bad IMO, considering the initial image set:

http://www.three60views.org.uk/texas_ca … _6k3k.html

Great!  also make sure you expose to the right and don't sharpen or otherwise post process.

Experiment with smaller apertures, like f/11 or f/13 or even f/16 -- you are balancing DOF and diffraction.  While the calculator I used said "big DOF", DOF is more complex and there's no easy answer.  Also, SIFT is by nature a "pixel peeper", so every pixel has to be good. 

Experiment with focus distance--
1. try getting the near edge of the DOF to about as high your lens is off the ground, or even 50% further than that.
2. try setting the focus distance to about 1/3 of the distance between you and the walls.

HDR may be useful here.  Meter outside for the f/stop you will use and remember the speed.  Take one circle with that setting to get the doors right.   Then shoot inside with AEB.  Use PM or other HDR program to merge the images, but lie to it--tell it the shot metered for outside is 0EV, and one of the AEB shots is 0EV.  Or use tufuse.

Last edited by hankkarl (2008-07-22 18:28:06)

davidweaver wrote:

Wow!  Great words of pano wisdom!
Fortunately I can reshoot this easily.
I'll try for sharper images.  This time I'll take a cable release. I know I should have done this but I wasn't actually planning on panos that day - I was killing time. :-).

I don't think the lack of sharpness is anything to do with not using a cable release/remote control. It is much more likely that the focus was set wrong on the Sigma 8mm. Are you using autofiocus? If so, don't. Use manual focus and don't set it on the infinity mark either. You'll find lots on the Panoguide.com forum via forum search about manually focusing the Sigma 8mm FE.

And do your best to get the expsoure right when shooting so there's no need to preprocess sthe images - are you shooting RAW or JPEG?

davidweaver wrote:

I'll shoot a set like this and another set at 15 degrees up..

Not sure 15 degrees is 'the magic number' - when I mentioned +15 I was thinking you were shooting with a Canon body, which doesn't produce a full 180 VFOV with the Sigma 8mm.

+5 will give a small overlap at the zenith on a Nikon DX body so I doubt you need as much as +15.

But anyway do try shooting the 'zenith' at an angle much less that +90. At + 60 you will get very good zenith coverage and the goal is to get an easier link between the 'zenith' shot and the main row - so try and take these lower angle 'zenith shots' where some readily identifiable feature will appear in the overlap of both a 'zenith' shot and a main row shot. I'd probably choose the doorway with the clock?

To get the feel/idea of this before trainling back to the State capitol building, just pop outside and shoot 4-around at +5 and one up at +60 and stitch it with APP - you can then see how it works and visualise the ovelaps using the layer tool in the Pano Editor by selecting all the images with Cnrtl-click. See screenshots below of a quick shoot in the garden using either two or one 'zenith' at +60, and 4-around main row at +5.

davidweaver wrote:

I shoot 8 around to give me more flexibility in keeping people in and out of frames.   I will post new new pics and results back here soon.

8-around if all stitched will have massive overlaps and could result in a lot of superfluous/spurious links with auto CP detection and which may reduce stitch accuracy. Or perhaps you meant 4-around twice? Smartblend usually does a pretty good job with 'ghosts' anyway.

You'll probably always have to set CPs manually with this sort of scene which has a lot of repetition of architectural features which will confuse most automatic CP detection systems.

Last edited by mediavets (2008-07-22 18:59:36)