jannemus / inertialscale Goto Github PK

how about detect features by de ep learning instead of using SIFT from visualSFM ??

Hi, thanks for the open-source code.
I am trying to use this code for scale alignment but did not get reasonable results.
I check the code and find there might be a bug in solveClosedForm function.
After the Umeyama alignment, we recover the Rs, but when rotating IMU measure to camera coordinate, e.g., this line, we should use angImu*Rs.transpose() rather than angImu*Rs .
After fixing this bug (actually, I use the unofficial C++ implementation, which is almost the same as this implementation), I can get the right extrinsic, which is consistent with other self-calibrated VIO systems, such as VINS-Mono. But I still can not get a reasonable scale estimate.
Can you give me some advice?

Hi,

There is currently no license to use your code, and I would appreciate one so I can check it out and use the algorithm.

Thank You!