Closed EchoTHChen closed 2 years ago
We converted each panorama into six cube-map images (120-degree FoV to make them overlap each other) that share the same camera position, then used them all.
I have another questions: For indoor scenes (there are many non-texture regions), the feature points extracted in the overlapped regions between cube-maps might be too few for COLMAP. So I think theoretically COLMAP might fail in some indoor scenes.
Could you share your scripts you generate the cube-maps and test on the sample data?(I just started learning panoramas recently and want to learn how COLMAP works for panorama?) Thanks!
I used py360convert. Please refer https://github.com/sunset1995/py360convert
As you can see in their document, we can convert them with e2p function as below.
fov=120 img = py360convert.e2p(frame, (fov, fov), -90, 0, (w,h), 0, 'bilinear')
Thanks!
I see that the repo of colmap does not provide the scripts of running colmap on panoramas:https://github.com/colmap/colmap/issues/790.
How can I run colmap on the panoramas (such as the dataset your method runs on)?