auto
: Automatic selection of best camera projection model.brown
: Robust rectilinear projection model. Preferred by auto
.fisheye
: Wide-angle / non-rectilinear projection model.perspective
: Legacy rectilinear projection model.spherical
: 360° camera projection model.auto
Resource | Impact |
---|---|
CPU | ●○○ | Low |
GPU | ○○○ | None |
HDD | ●○○ | Low |
RAM | ●○○ | Low |
Time | ●○○ | Low |
Camera Lens Models are projection/distortion models that OpenSFM uses to correct for the optics of the camera platforms that record our images. These corrections are essential for proper camera/scene modeling, and therefore, proper reconstruction of the data.
Manually choosing a Camera Lens Model is currently only recommended in the case of Spherical Cameras (GoPro Fusion, GoPro Max, Insta360, Kodak PixPro, etc.) as these are currently not detected automatically by OpenSFM.
In rare cases, OpenSFM may not be able to detect (or retrieve from its Camera database) the correct Camera Lens Model for your particular sensor, in which case you should select the closest appropriate model. When in doubt, try specifying brown
first.
The following examples are all data taken with a standard Rectilinear Lens. Manual selection of the various Camera Lens Models is demonstrated to show some differences in how this will affect reconstruction.
This Point Cloud shows a proper reconstruction via the auto
(brown
) Camera Lens Model, which is appropriate for this sensor.
In this Point Cloud reconstruction, severe bowling and other artifacts have been introduced via the use of the wrong Camera Lens Model.
This Point Cloud rendered well, very similar to the auto
and brown
examples. One may see artifacts in the perspective
Camera Lens Model in very large collections, or over very flat/homogenus terrain (agricultural fields). In such cases, forcing brown
may help.
This Point Cloud failed to reconstruct properly due to the manual selection of the fisheye
Camera Lens Model. Similar failures to reconstruct can often indicate the wrong manual selection, or in edge cases, wrong auto
selection of the Camera Lens Model, and you are advised to try another Model that is more appropriate.