Optimizing problems

Optimizing problems

The value displayed during optimisation is somewhat too high

First note that this value represents the average distance between matching contral points in the resulting panorama.Thus for a given set of images and control points, it is proportionnal to the size of the panorama. So 2 pixels distance for a 1000x500 panorama is bad, 3 pixels for a 3000x1500 panorama is correct (but not outstanding !)

You didn't rotate your camera around its nodal point, and you got parallax errors (for more information, see Panoguide article on parallax, and Coolpix nodal point positions at 360texas.com). This problem is usually more severe in tight places than in landscape panoramas. If you can't re-shoot your panorama, try to minimize the problem by selecting control points in faraway features. Seams in nearby objects will be bad though.
One (or more) of your control point is bad. Go to the Control points tab (after applying optimizing results). The distance between each matching control points will be displayed. Try to delete the control points with the highest error and replace them if necessary. Tip: since you applied (wrong) optimizing results to your startup values, this can prevent convergence of the optimizer after you removed bad control points. In that case, reload your script (you didn't forget to save it before optimization, did you ?) and delete those bad C.P. before any optimization.

The value displayed during optimisation is much too high and/or the hfov is obviously wrong for my lens

A few possibilities...

You selected a template that doesn't match your situation
Your images are not in the right order. Remember that my templates requires that you name the images in alphanumeric order as your camera is rotated clockwise.
Some control points are very wrong.
You try to optimize too many parameters and your control points don't provide informations that the optimizer can use to compute them (this should not happen with the templates I provide...). One solution (obviously wrong, but mathematically correct) to minimize distance between matching control points is to reduce each image to a point and overlap all the images : the distance will be zero. This case can happen for example if you try to optimize for hfov and distortion parameters (a, b and c) with a two fisheye images 180° apart. Since all control points you can provide are located approximately at the same distance of images centers, the optimizer has no way to guess these values and will probably converge towards all 0.



		Optimizing problems The value displayed during optimisation is somewhat too high First note that this value represents the average distance between matching contral points in the resulting panorama.Thus for a given set of images and control points, it is proportionnal to the size of the panorama. So 2 pixels distance for a 1000x500 panorama is bad, 3 pixels for a 3000x1500 panorama is correct (but not outstanding !) You didn't rotate your camera around its nodal point, and you got parallax errors (for more information, see Panoguide article on parallax, and Coolpix nodal point positions at 360texas.com). This problem is usually more severe in tight places than in landscape panoramas. If you can't re-shoot your panorama, try to minimize the problem by selecting control points in faraway features. Seams in nearby objects will be bad though. One (or more) of your control point is bad. Go to the Control points tab (after applying optimizing results). The distance between each matching control points will be displayed. Try to delete the control points with the highest error and replace them if necessary. Tip: since you applied (wrong) optimizing results to your startup values, this can prevent convergence of the optimizer after you removed bad control points. In that case, reload your script (you didn't forget to save it before optimization, did you ?) and delete those bad C.P. before any optimization. The value displayed during optimisation is much too high and/or the hfov is obviously wrong for my lens A few possibilities... You selected a template that doesn't match your situation Your images are not in the right order. Remember that my templates requires that you name the images in alphanumeric order as your camera is rotated clockwise. Some control points are very wrong. You try to optimize too many parameters and your control points don't provide informations that the optimizer can use to compute them (this should not happen with the templates I provide...). One solution (obviously wrong, but mathematically correct) to minimize distance between matching control points is to reduce each image to a point and overlap all the images : the distance will be zero. This case can happen for example if you try to optimize for hfov and distortion parameters (a, b and c) with a two fisheye images 180° apart. Since all control points you can provide are located approximately at the same distance of images centers, the optimizer has no way to guess these values and will probably converge towards all 0.