colmap2nerf: Use separate intrinsics if available #1357
Conversation
…amera intrinsics per frame
scripts/colmap2nerf.py
Outdated
| is_fisheye = False | ||
| camera = {} | ||
| camera_id = int(els[0]) | ||
| camera['w'] = float(els[2]) |
There was a problem hiding this comment.
Could all the ''-strings be changed to ""-strings?
scripts/colmap2nerf.py
Outdated
| @@ -333,6 +348,23 @@ def closest_point_2_lines(oa, da, ob, db): # returns point closest to both rays | |||
| up += c2w[0:3,1] | |||
|
|
|||
| frame = {"file_path":name,"sharpness":b,"transform_matrix": c2w} | |||
| if len(cameras) != 1: | |||
| camera = cameras[int(elems[8])] | |||
| frame["camera_angle_x"] = camera["angle_x"] | |||
There was a problem hiding this comment.
This could be expressed more concisely using frame |= camera if the angle_x parameter is renamed to camera_angle_x.
|
Sorry it took a while, but I applied the suggested changes.
|
|
Oh, I hadn't realized |
…ate() instead of |-
|
I think we can get the best of both worlds by using Also I accidentally changed |
|
LGTM, thanks again! |
colmap2nerf: Use separate intrinsics if available
Currently only the last images' intrinsics are being used.
Recently COLMAP added a checkbox to do automatic reconstruction with separate intrinsics per subfolder.
I'm using a rig with multiple cameras (same model), so I thought it would be nice to see if using intrinsics per camera would improve the reconstruction.
Actually there doesn't seem to be a qualitative improvement, but maybe my modification to the script is still useful when for example using non-homogeneous cameras/lenses.