Is the reconstructed model in real-world scale?

[CanCanZeng]

Hi, thank you for sharing this code! I find that this is the only one open sourced project that try to do SfM with the aid of imu, does it just like visult-inertial odometry that reconstruct the scene with real-world scale? Or you just use the gyroscope to predict the rotation of next image?

[hovren]

If you have accelerometer measurements the reconstructed scene is at real-world scale. Den tors 11 aug. 2022 15:20CanCanZeng ***@***.***> skrev:

…

Hi, thank you for sharing this code! I find that this is the only one open sourced project that try to do SfM with the aid of imu, does it just like visult-inertial odometry that reconstruct the scene with real-world scale? Or you just use the gyroscope to predict the rotation of next image? — Reply to this email directly, view it on GitHub <#4>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACMB35LLT46D6SGHODZCX3VYT4Z3ANCNFSM56IEUL3A> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[CanCanZeng]

Thanks for your reply, glad to hear that! Do you estimates the bias of gyro and acce data when solving the sfm problem? It seems that if the time interval between adjecent images is large(for example a few seconds), it is very hard to use the imu info since the bias of imu will break the system.

And can you provide some example data?