You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Mar 9, 2026. It is now read-only.
Repository navigation
This repository was archived by the owner on Mar 9, 2026. It is now read-only.
I encountered a problem when using glomap for SfM with a fisheye camera. I used the OPENCV_FISHEYE camera model and the latest code from the glomap main branch. When I run glomapper on features extracted with OPENCV_FISHEYE, the result appears as shown in the image below.
However, when I convert the fisheye camera images into four pinhole images, glomapper works very well. Can you provide some insights on this issue?
Hi @s1752729916 , sorry for the late reply. There can be several potential reasons for the failure: matches cannot be reliably verifed by two-view geometry, camera intrinsics estimation failed, fisheye cameras are difficult to calibrate in general, etc. To rule out the potential failure from GLOMAP itself, you can try to first calibrate the fisheye images with a few images with COLMAP, then feed in the calibrated parameters to GLOMAP and run again. It should give you reasonable results from my understanding.
Activity
lpanaf commentedon Jan 12, 2026
Hi @s1752729916 , sorry for the late reply. There can be several potential reasons for the failure: matches cannot be reliably verifed by two-view geometry, camera intrinsics estimation failed, fisheye cameras are difficult to calibrate in general, etc. To rule out the potential failure from GLOMAP itself, you can try to first calibrate the fisheye images with a few images with COLMAP, then feed in the calibrated parameters to GLOMAP and run again. It should give you reasonable results from my understanding.