Forum: Q & A

If your lens parameters are set correctly, and the nodal point is used, there should be no problem

Hans Nyberg said:

I must say that it surprices me that you can get any stitch at all from AutopanoPro. I really would like to see your panorama.

I got a trial yesterday and made some tests.
I tested 3 types of fisheye panos.
2 fullframe set of 6+1 zenith
3 different Sigma 8mm with 4 or 4+1.
All perfect setups on panohead which Ptgui stitches perfect in a couple of minutes.

AutopanoPro could not do anything that looked like a panorama from any of them,

Hans Nyberg said:

I have used and tested almost every stitcher on the market the last 10 years and I have to say that I never seen anything like this.

Hans Nyberg said:

At last I found a simple cylindric pano and it turned out that this worked perfect.

Hans Nyberg said:

What I could see was that Autopano detected all lensdata perfect for the fisheyes and also set the crop perfect automatically.

Hans Nyberg said:

But the controlpoints was very bad. Sometimes it even found controlpoints between images which had no overlap at all.
For example on the 4 around I got controlpoints between 0 and 180
And it also detected lots of them on the rotator which of course screwed it all up.

Hans Nyberg said:

It is extremely slow and just a simple thing like centering the image in the editor takes 10 sec, PTgui reacts instantly and rearranges the images in 1-2 sec

Hans Nyberg said:

The whole interface is very confusing and it uses different definitions for the same thing. Controlpoints is called both Keypoints and Controlpoints

Hans Nyberg said:

The default controlpoint setting is to generate 50 per image pair which is just insane. You should not use more than 15.

John Houghton said:

My tests with Autopano Pro produced mixed results. I started off with some images taken with a Nikon 10.5mm fisheye on a Canon camera. The results were very poor, probably due mainly to the fact that the exif data did not contain information about the lens. It therefore treated it as rectilinear. I looked hard for help on how to specify the lens details. Pressing Help provided no help. I went to the web site and a text search suggested that I might be able to do this under "Settings", but I couldn't find any appropriate places there. So that project was abandoned.

John Houghton said:

I then tried a stitch of images from a Sigma 8mm f3.5: 4 around + Z + N. The nadir was taken on the tripod but to one side. PTGui Pro produces a perfect stitch using the viewpoint correction feature, though some manual control is needed. Autopano Pro stitched the nadir poorly, but the horizontals and zenith were ok. No doubt a handheld shot taken nearer to the original camera position would have produced a better result.

John Houghton said:

As Pierre says, the Autopano Pro image was very unsharp compared to the PTGui output. I always sharpen the final image anyway, and the Autopano Pro image sharpened well enough in Photoshop, becoming very little different to that from PTGui.

mediavets said:

As they say 'if all else fails RTFM'.

John Houghton said:

mediavets said:

As they say 'if all else fails RTFM'.

John Houghton said:

Andrew, Thank you for that sensitively expressed advice.

Was meant tongue in cheek. It's a commonly used light hearted acronym in the IT industry in which I used to work. And so many people don't.

John Houghton said:

Generally good results were obtained but there were some minor problems. I sometimes got a bad stitch for no obvious reason, but a good one when I repeated it.

When you got a 'bad stitch' did you check with the Control Points Editor to see whether there were bad links or high value CPs?

John Houghton said:

I tried to add in a nadir image but was told that mixed lens types were not supported. (It was the same as all the other images). I couldn't select that one image and change its type to fisheye, so I had to start again from scratch.

You might have forced the nadir image into the pano to start with - even if APP does not automatically detect CPs - there is an option to do that. I tend to choose that if pano includes lots of featureless sky shots, for example.

In absence of lens type data in EXIFs APP doesn't try and guess whether a lens is FE or rectilinear based on based on focal length - some other stitchers do. One needs to check they are all set to the appropriate type in image properties before stitching.

John Houghton said:

Some images had alpha channel masks, but the stitched output showed black lines corresponding to the mask and image edges.

What was the purpose of the masks?
Which interpolator and blend type did you select when rendering?

I'm sure a lot of people will find the results from Autopano Pro more than acceptable, especially when a routine workflow has been established. I am not tempted to switch from PTGui just yet though. There's no reason to restrict oneself exclusively to PTGui OR Autopan PRO. Why not have both?

But if you're shooting the nadir oblique, then it doesnt suprise me that you wont get a stitch.....

John Houghton said:

Andrew, You should be appreciate that while most people will be amused, others will be offended by RTFM - once they know what the letters stand for.

John Houghton said:

My poor stitches may indeed have been due to bad control points. I'm afraid I quickly lost patience with the control point editing facilities. In this area, PTGui seems far more straightforward. There seems to be no way of correcting control points or adding them manually. (I could find no mention of doing these things in the manual).

John Houghton said:

Masks are very useful for eliminating parts of an image from the blending process. Double images caused by movement in overlaps can be avoided by masking one or other out of the input images. Intrusive bits of pano head/tripod hardware and photographer's feet can also be eliminated too. If left in, they can often cause odd blending shadows. It is often easier to deal with these problems by editing the input images rather than editing with layers in postprocessing. I left the interpolator and blending choices at their default values of bilinear and multiband. These should have no effect on the image blending.