Hello,
I have a sfm algorithm that gives me the features and camera extrinsics, and i have already tried writing my own synth_0.out file and tried running MVE from the dmrecon, but i only manage to get global view selection errors, do you have any ideas on where i might be doing something wrong?
Best regards.
I recommend running the MVE SfM (e.g., on a small subset of the images) just to see what the output is. For example, you need to put your intrinsics and extrinsics in the meta.ini
file for each view. synth_0.out
is only used to describe the tracks, but not for intrinsics/extrinsics. Let me know if you have any questions.
Also, if you're using a popular tool (bundler, VisualSfM) for your SfM reconstruction, makescene
can actually read these formats.
Sorry, i forget to say that i also write the meta.ini file for each view, and i also did the undistorted image file.
Then you have to give me some more data points. Please post the dmrecon
error message you're getting, show me one of the meta.ini
files, ls
one of your view directories, and make sure the scene shows up properly in UMVE.
MVE_terminal_dmreconFail.txt
This is the dmrecon error message
view_0044.mve.zip
This is one of my view directories with the meta.ini file inside it
synth_0.zip
This is my synth_0.out file
And i have already tried running this same scene directory with the full MVE pipeline, and it works just fine, so i have excluded problems with the dataset, probably something arround the synth file.
I am pretty certain there is something wrong with your bundle file synth_0.out
. For example, your point visibilities look like this:
3 23 0 0 53 0 0 56 0 0
and the format is
<num observations> <img id> <sift id> <reproj. error> ...
where all of your SIFT IDs are 0
. The reprojection error is indeed ignored, but I the SIFT IDs are important.
I will look more into it, but i tried using 0 in the SIFT IDs with the MVE complete pipeline and it worked fine, so i thought that it wasn't important.
Actually you might be right, feature IDs are not important after bundle adjustment. Then maybe your coordinate systems are off, i.e., the 3D feature points don't properly project back into the images. It could be an issue with the 3D feature coordinate system, or with the extrinsic parameters. An easy way to check that is to use UMVE and check your scene visually.
The camera transformation matrix in the synth_0 file is the world to camera matrix or the camera to world matrix?
It's the world to camera transform.
Closing due to inactivity. Feel free to reopen.
Most helpful comment
I recommend running the MVE SfM (e.g., on a small subset of the images) just to see what the output is. For example, you need to put your intrinsics and extrinsics in the
meta.ini
file for each view.synth_0.out
is only used to describe the tracks, but not for intrinsics/extrinsics. Let me know if you have any questions.Also, if you're using a popular tool (bundler, VisualSfM) for your SfM reconstruction,
makescene
can actually read these formats.