Crop and FOV issue with circular fisheye

I found this post - https://groups.google.com/forum/#!topic/hugin-ptx/dkseRDM8X30 - which is about circular fisheye FOV, but it doesn't solve any of my issues.

In Hugin the fov refers to the horizontal field of view of the uncropped image. PTGui is using another convention. So the parameter can not exchanged 1:1 between Hugin and PTGui.
See http://wiki.panotools.org/Panotools_internals#Cropping

The value of 162 is too low. So maybe there are some wrong control points and the optimizer has gone wrong.

PS: The projection of the Yasuhara Madoka 180 is orthographic (http://wiki.panotools.org/Fisheye ). But this projection make problems when using a cropped image in landscape.
I see currently 3 workarounds
1) Rotate the images before feeding into Hugin to portrait. Then the hfov becomes smaller and can be better handled by the optimizer.
2) Crop the image to the crop circle outside Hugin (use a square image)
3) Use circular fisheye as image projection. But then probably higher values for a/b/c are needed.

Thanks for the info.

I tried rotating the photo to portrait beforehand (as a png) and then setting the lens type to Orthographic with a focal length of 7.3mm and crop factor of 2, and left the lens parameters at zero.

That results in a calculated FOV of 125.51 degrees.

However, after cropping to the image circle I still get an output image with the FOV narrower than it should be.

https://www.dropbox.com/s/ox7mp38ynovt63p/Hugin_PreRotated_Manual7.3mmFocalLenth_Orthogonal_Zero_ABC_values.jpg?dl=0

Are you able to take the following original (rotated) photo and correctly convert it to equirectangular?

https://www.dropbox.com/s/r9yq617bwu0357n/OriginalPhoto_Rotated.JPG?dl=0

Any further insight would be appreciated.

Thanks,
Dustin

> However, after cropping to the image circle I still get an output image with the FOV narrower than it should be.

I don't know what you expect. The image expands 180 degree vertically. And horizontally it expands also 180 degrees (at least on top and bottom of the equirectangular). The missing (white) half circles in the middle are not in the posted original image. (The image circle is slightly larger than the smaller image dimensions. So there are some little parts missing from the full half sphere.)
Hugin can't invent image content which is not in the original images.

Sorry for not being clearer. I'm not talking about the missing circular regions. The photo is indeed a circular fisheye image partially cropped by the sensor. My issue is the missing regions at the extreme edges all around the image (top, bottom, and left/right beyond the missing half circles).

Here is an easy to view comparison between the two converted 180x180 Equirectangulars from PTGui (on the left) and Hugin (on the right) with red circles on some of the regions I'm talking about:

https://www.dropbox.com/s/wt8vza4iko0rpz0/ComparisonPTgui_Hugin.jpg?dl=0

These photos were converted using the following steps for both programs:

1. Rotate the photo to portrait in Photoshop
2. Load the image in PTGui/Hugin
3. Set the lens type to Orthographic (Hugin calculates FOV=125.51) and Circular Fisheye (PTGui calculates FOV=180) with a focal length of 7.3mm and crop factor of 2
4. Use the circle crop tool to crop to the image area
5. Reset lens parameters to default (zero)
6. Set output to Equirectangular 180x180

While neither of these output images are going to be perfect since I didn't use lens parameters, you can easily see that the Hugin output is not including the entire image area.

With this pre-rotated image, I also tried changing the default calculated Orthographic FOV in Hugin from 125.51 -> 180 and also using Circular Fisheye in Hugin with both the default calculated 101.88 FOV and manually set 180 FOV but none of those look like my PTGui image. The only thing that gets me close in Hugin is to use Circular Fisheye with a manually set FOV of around 150 at a crop factor of 2, which is the behavior I noted in my first post (though the numbers are different with the pre-rotated image).

Does that make sense? Would you be able to try converting that pre-rotated image yourself? It's very possible I'm just using Hugin incorrectly.

Thanks,
Dustin

Orthographic and fisheye use different mapping function. So you get different results especially and the image borders. And at the image border the results depends strongly on the fov (the printed focal length of the lens is too vague and the projection which uses the manufacturer to calculate the focal length is unknown, so the focal length is not precise enough for this task).

With your prerotated image and and fov of 118 I'm getting more at the image borders (or also with fisheye circular and a fov of around 150).
But instead of fiddeling with a single photo, shot a full circle with several images (e.g. 4 images) and try to stitch the images. Then you will get the fov (and maybe also the distortion parameters a, b, c) how Hugin expects its. In this scenario you can try both projections and see which one matches better with your lens.

Sounds like you're getting similar results to what I saw.

From my initial test with 10 photos I was seeing this same level of over cropping, but I'll give it another shot with a new set of photos. I'll post back once I get that completed.

-Dustin